2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE = 0,
51 CHUNK_ALLOC_FORCE = 1,
52 CHUNK_ALLOC_LIMITED = 2,
55 static int update_block_group(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 bytenr, u64 num_bytes, int alloc);
58 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, u64 parent,
61 u64 root_objectid, u64 owner_objectid,
62 u64 owner_offset, int refs_to_drop,
63 struct btrfs_delayed_extent_op *extra_op);
64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
65 struct extent_buffer *leaf,
66 struct btrfs_extent_item *ei);
67 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_root *root,
69 u64 parent, u64 root_objectid,
70 u64 flags, u64 owner, u64 offset,
71 struct btrfs_key *ins, int ref_mod);
72 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
73 struct btrfs_root *root,
74 u64 parent, u64 root_objectid,
75 u64 flags, struct btrfs_disk_key *key,
76 int level, struct btrfs_key *ins);
77 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
78 struct btrfs_root *extent_root, u64 alloc_bytes,
79 u64 flags, int force);
80 static int find_next_key(struct btrfs_path *path, int level,
81 struct btrfs_key *key);
82 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
83 int dump_block_groups);
86 block_group_cache_done(struct btrfs_block_group_cache *cache)
89 return cache->cached == BTRFS_CACHE_FINISHED;
92 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
94 return (cache->flags & bits) == bits;
97 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
99 atomic_inc(&cache->count);
102 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
104 if (atomic_dec_and_test(&cache->count)) {
105 WARN_ON(cache->pinned > 0);
106 WARN_ON(cache->reserved > 0);
107 WARN_ON(cache->reserved_pinned > 0);
113 * this adds the block group to the fs_info rb tree for the block group
116 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
117 struct btrfs_block_group_cache *block_group)
120 struct rb_node *parent = NULL;
121 struct btrfs_block_group_cache *cache;
123 spin_lock(&info->block_group_cache_lock);
124 p = &info->block_group_cache_tree.rb_node;
128 cache = rb_entry(parent, struct btrfs_block_group_cache,
130 if (block_group->key.objectid < cache->key.objectid) {
132 } else if (block_group->key.objectid > cache->key.objectid) {
135 spin_unlock(&info->block_group_cache_lock);
140 rb_link_node(&block_group->cache_node, parent, p);
141 rb_insert_color(&block_group->cache_node,
142 &info->block_group_cache_tree);
143 spin_unlock(&info->block_group_cache_lock);
149 * This will return the block group at or after bytenr if contains is 0, else
150 * it will return the block group that contains the bytenr
152 static struct btrfs_block_group_cache *
153 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
156 struct btrfs_block_group_cache *cache, *ret = NULL;
160 spin_lock(&info->block_group_cache_lock);
161 n = info->block_group_cache_tree.rb_node;
164 cache = rb_entry(n, struct btrfs_block_group_cache,
166 end = cache->key.objectid + cache->key.offset - 1;
167 start = cache->key.objectid;
169 if (bytenr < start) {
170 if (!contains && (!ret || start < ret->key.objectid))
173 } else if (bytenr > start) {
174 if (contains && bytenr <= end) {
185 btrfs_get_block_group(ret);
186 spin_unlock(&info->block_group_cache_lock);
191 static int add_excluded_extent(struct btrfs_root *root,
192 u64 start, u64 num_bytes)
194 u64 end = start + num_bytes - 1;
195 set_extent_bits(&root->fs_info->freed_extents[0],
196 start, end, EXTENT_UPTODATE, GFP_NOFS);
197 set_extent_bits(&root->fs_info->freed_extents[1],
198 start, end, EXTENT_UPTODATE, GFP_NOFS);
202 static void free_excluded_extents(struct btrfs_root *root,
203 struct btrfs_block_group_cache *cache)
207 start = cache->key.objectid;
208 end = start + cache->key.offset - 1;
210 clear_extent_bits(&root->fs_info->freed_extents[0],
211 start, end, EXTENT_UPTODATE, GFP_NOFS);
212 clear_extent_bits(&root->fs_info->freed_extents[1],
213 start, end, EXTENT_UPTODATE, GFP_NOFS);
216 static int exclude_super_stripes(struct btrfs_root *root,
217 struct btrfs_block_group_cache *cache)
224 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
225 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
226 cache->bytes_super += stripe_len;
227 ret = add_excluded_extent(root, cache->key.objectid,
232 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
233 bytenr = btrfs_sb_offset(i);
234 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
235 cache->key.objectid, bytenr,
236 0, &logical, &nr, &stripe_len);
240 cache->bytes_super += stripe_len;
241 ret = add_excluded_extent(root, logical[nr],
251 static struct btrfs_caching_control *
252 get_caching_control(struct btrfs_block_group_cache *cache)
254 struct btrfs_caching_control *ctl;
256 spin_lock(&cache->lock);
257 if (cache->cached != BTRFS_CACHE_STARTED) {
258 spin_unlock(&cache->lock);
262 /* We're loading it the fast way, so we don't have a caching_ctl. */
263 if (!cache->caching_ctl) {
264 spin_unlock(&cache->lock);
268 ctl = cache->caching_ctl;
269 atomic_inc(&ctl->count);
270 spin_unlock(&cache->lock);
274 static void put_caching_control(struct btrfs_caching_control *ctl)
276 if (atomic_dec_and_test(&ctl->count))
281 * this is only called by cache_block_group, since we could have freed extents
282 * we need to check the pinned_extents for any extents that can't be used yet
283 * since their free space will be released as soon as the transaction commits.
285 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
286 struct btrfs_fs_info *info, u64 start, u64 end)
288 u64 extent_start, extent_end, size, total_added = 0;
291 while (start < end) {
292 ret = find_first_extent_bit(info->pinned_extents, start,
293 &extent_start, &extent_end,
294 EXTENT_DIRTY | EXTENT_UPTODATE);
298 if (extent_start <= start) {
299 start = extent_end + 1;
300 } else if (extent_start > start && extent_start < end) {
301 size = extent_start - start;
303 ret = btrfs_add_free_space(block_group, start,
306 start = extent_end + 1;
315 ret = btrfs_add_free_space(block_group, start, size);
322 static int caching_kthread(void *data)
324 struct btrfs_block_group_cache *block_group = data;
325 struct btrfs_fs_info *fs_info = block_group->fs_info;
326 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
327 struct btrfs_root *extent_root = fs_info->extent_root;
328 struct btrfs_path *path;
329 struct extent_buffer *leaf;
330 struct btrfs_key key;
336 path = btrfs_alloc_path();
340 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
343 * We don't want to deadlock with somebody trying to allocate a new
344 * extent for the extent root while also trying to search the extent
345 * root to add free space. So we skip locking and search the commit
346 * root, since its read-only
348 path->skip_locking = 1;
349 path->search_commit_root = 1;
354 key.type = BTRFS_EXTENT_ITEM_KEY;
356 mutex_lock(&caching_ctl->mutex);
357 /* need to make sure the commit_root doesn't disappear */
358 down_read(&fs_info->extent_commit_sem);
360 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
364 leaf = path->nodes[0];
365 nritems = btrfs_header_nritems(leaf);
369 if (fs_info->closing > 1) {
374 if (path->slots[0] < nritems) {
375 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
377 ret = find_next_key(path, 0, &key);
381 if (need_resched() ||
382 btrfs_next_leaf(extent_root, path)) {
383 caching_ctl->progress = last;
384 btrfs_release_path(extent_root, path);
385 up_read(&fs_info->extent_commit_sem);
386 mutex_unlock(&caching_ctl->mutex);
390 leaf = path->nodes[0];
391 nritems = btrfs_header_nritems(leaf);
395 if (key.objectid < block_group->key.objectid) {
400 if (key.objectid >= block_group->key.objectid +
401 block_group->key.offset)
404 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
405 total_found += add_new_free_space(block_group,
408 last = key.objectid + key.offset;
410 if (total_found > (1024 * 1024 * 2)) {
412 wake_up(&caching_ctl->wait);
419 total_found += add_new_free_space(block_group, fs_info, last,
420 block_group->key.objectid +
421 block_group->key.offset);
422 caching_ctl->progress = (u64)-1;
424 spin_lock(&block_group->lock);
425 block_group->caching_ctl = NULL;
426 block_group->cached = BTRFS_CACHE_FINISHED;
427 spin_unlock(&block_group->lock);
430 btrfs_free_path(path);
431 up_read(&fs_info->extent_commit_sem);
433 free_excluded_extents(extent_root, block_group);
435 mutex_unlock(&caching_ctl->mutex);
436 wake_up(&caching_ctl->wait);
438 put_caching_control(caching_ctl);
439 atomic_dec(&block_group->space_info->caching_threads);
440 btrfs_put_block_group(block_group);
445 static int cache_block_group(struct btrfs_block_group_cache *cache,
446 struct btrfs_trans_handle *trans,
447 struct btrfs_root *root,
450 struct btrfs_fs_info *fs_info = cache->fs_info;
451 struct btrfs_caching_control *caching_ctl;
452 struct task_struct *tsk;
456 if (cache->cached != BTRFS_CACHE_NO)
460 * We can't do the read from on-disk cache during a commit since we need
461 * to have the normal tree locking. Also if we are currently trying to
462 * allocate blocks for the tree root we can't do the fast caching since
463 * we likely hold important locks.
465 if (trans && (!trans->transaction->in_commit) &&
466 (root && root != root->fs_info->tree_root)) {
467 spin_lock(&cache->lock);
468 if (cache->cached != BTRFS_CACHE_NO) {
469 spin_unlock(&cache->lock);
472 cache->cached = BTRFS_CACHE_STARTED;
473 spin_unlock(&cache->lock);
475 ret = load_free_space_cache(fs_info, cache);
477 spin_lock(&cache->lock);
479 cache->cached = BTRFS_CACHE_FINISHED;
480 cache->last_byte_to_unpin = (u64)-1;
482 cache->cached = BTRFS_CACHE_NO;
484 spin_unlock(&cache->lock);
486 free_excluded_extents(fs_info->extent_root, cache);
494 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
495 BUG_ON(!caching_ctl);
497 INIT_LIST_HEAD(&caching_ctl->list);
498 mutex_init(&caching_ctl->mutex);
499 init_waitqueue_head(&caching_ctl->wait);
500 caching_ctl->block_group = cache;
501 caching_ctl->progress = cache->key.objectid;
502 /* one for caching kthread, one for caching block group list */
503 atomic_set(&caching_ctl->count, 2);
505 spin_lock(&cache->lock);
506 if (cache->cached != BTRFS_CACHE_NO) {
507 spin_unlock(&cache->lock);
511 cache->caching_ctl = caching_ctl;
512 cache->cached = BTRFS_CACHE_STARTED;
513 spin_unlock(&cache->lock);
515 down_write(&fs_info->extent_commit_sem);
516 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
517 up_write(&fs_info->extent_commit_sem);
519 atomic_inc(&cache->space_info->caching_threads);
520 btrfs_get_block_group(cache);
522 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
523 cache->key.objectid);
526 printk(KERN_ERR "error running thread %d\n", ret);
534 * return the block group that starts at or after bytenr
536 static struct btrfs_block_group_cache *
537 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
539 struct btrfs_block_group_cache *cache;
541 cache = block_group_cache_tree_search(info, bytenr, 0);
547 * return the block group that contains the given bytenr
549 struct btrfs_block_group_cache *btrfs_lookup_block_group(
550 struct btrfs_fs_info *info,
553 struct btrfs_block_group_cache *cache;
555 cache = block_group_cache_tree_search(info, bytenr, 1);
560 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
563 struct list_head *head = &info->space_info;
564 struct btrfs_space_info *found;
566 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
567 BTRFS_BLOCK_GROUP_METADATA;
570 list_for_each_entry_rcu(found, head, list) {
571 if (found->flags & flags) {
581 * after adding space to the filesystem, we need to clear the full flags
582 * on all the space infos.
584 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
586 struct list_head *head = &info->space_info;
587 struct btrfs_space_info *found;
590 list_for_each_entry_rcu(found, head, list)
595 static u64 div_factor(u64 num, int factor)
604 static u64 div_factor_fine(u64 num, int factor)
613 u64 btrfs_find_block_group(struct btrfs_root *root,
614 u64 search_start, u64 search_hint, int owner)
616 struct btrfs_block_group_cache *cache;
618 u64 last = max(search_hint, search_start);
625 cache = btrfs_lookup_first_block_group(root->fs_info, last);
629 spin_lock(&cache->lock);
630 last = cache->key.objectid + cache->key.offset;
631 used = btrfs_block_group_used(&cache->item);
633 if ((full_search || !cache->ro) &&
634 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
635 if (used + cache->pinned + cache->reserved <
636 div_factor(cache->key.offset, factor)) {
637 group_start = cache->key.objectid;
638 spin_unlock(&cache->lock);
639 btrfs_put_block_group(cache);
643 spin_unlock(&cache->lock);
644 btrfs_put_block_group(cache);
652 if (!full_search && factor < 10) {
662 /* simple helper to search for an existing extent at a given offset */
663 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
666 struct btrfs_key key;
667 struct btrfs_path *path;
669 path = btrfs_alloc_path();
671 key.objectid = start;
673 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
674 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
676 btrfs_free_path(path);
681 * helper function to lookup reference count and flags of extent.
683 * the head node for delayed ref is used to store the sum of all the
684 * reference count modifications queued up in the rbtree. the head
685 * node may also store the extent flags to set. This way you can check
686 * to see what the reference count and extent flags would be if all of
687 * the delayed refs are not processed.
689 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
690 struct btrfs_root *root, u64 bytenr,
691 u64 num_bytes, u64 *refs, u64 *flags)
693 struct btrfs_delayed_ref_head *head;
694 struct btrfs_delayed_ref_root *delayed_refs;
695 struct btrfs_path *path;
696 struct btrfs_extent_item *ei;
697 struct extent_buffer *leaf;
698 struct btrfs_key key;
704 path = btrfs_alloc_path();
708 key.objectid = bytenr;
709 key.type = BTRFS_EXTENT_ITEM_KEY;
710 key.offset = num_bytes;
712 path->skip_locking = 1;
713 path->search_commit_root = 1;
716 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
722 leaf = path->nodes[0];
723 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
724 if (item_size >= sizeof(*ei)) {
725 ei = btrfs_item_ptr(leaf, path->slots[0],
726 struct btrfs_extent_item);
727 num_refs = btrfs_extent_refs(leaf, ei);
728 extent_flags = btrfs_extent_flags(leaf, ei);
730 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
731 struct btrfs_extent_item_v0 *ei0;
732 BUG_ON(item_size != sizeof(*ei0));
733 ei0 = btrfs_item_ptr(leaf, path->slots[0],
734 struct btrfs_extent_item_v0);
735 num_refs = btrfs_extent_refs_v0(leaf, ei0);
736 /* FIXME: this isn't correct for data */
737 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
742 BUG_ON(num_refs == 0);
752 delayed_refs = &trans->transaction->delayed_refs;
753 spin_lock(&delayed_refs->lock);
754 head = btrfs_find_delayed_ref_head(trans, bytenr);
756 if (!mutex_trylock(&head->mutex)) {
757 atomic_inc(&head->node.refs);
758 spin_unlock(&delayed_refs->lock);
760 btrfs_release_path(root->fs_info->extent_root, path);
762 mutex_lock(&head->mutex);
763 mutex_unlock(&head->mutex);
764 btrfs_put_delayed_ref(&head->node);
767 if (head->extent_op && head->extent_op->update_flags)
768 extent_flags |= head->extent_op->flags_to_set;
770 BUG_ON(num_refs == 0);
772 num_refs += head->node.ref_mod;
773 mutex_unlock(&head->mutex);
775 spin_unlock(&delayed_refs->lock);
777 WARN_ON(num_refs == 0);
781 *flags = extent_flags;
783 btrfs_free_path(path);
788 * Back reference rules. Back refs have three main goals:
790 * 1) differentiate between all holders of references to an extent so that
791 * when a reference is dropped we can make sure it was a valid reference
792 * before freeing the extent.
794 * 2) Provide enough information to quickly find the holders of an extent
795 * if we notice a given block is corrupted or bad.
797 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
798 * maintenance. This is actually the same as #2, but with a slightly
799 * different use case.
801 * There are two kinds of back refs. The implicit back refs is optimized
802 * for pointers in non-shared tree blocks. For a given pointer in a block,
803 * back refs of this kind provide information about the block's owner tree
804 * and the pointer's key. These information allow us to find the block by
805 * b-tree searching. The full back refs is for pointers in tree blocks not
806 * referenced by their owner trees. The location of tree block is recorded
807 * in the back refs. Actually the full back refs is generic, and can be
808 * used in all cases the implicit back refs is used. The major shortcoming
809 * of the full back refs is its overhead. Every time a tree block gets
810 * COWed, we have to update back refs entry for all pointers in it.
812 * For a newly allocated tree block, we use implicit back refs for
813 * pointers in it. This means most tree related operations only involve
814 * implicit back refs. For a tree block created in old transaction, the
815 * only way to drop a reference to it is COW it. So we can detect the
816 * event that tree block loses its owner tree's reference and do the
817 * back refs conversion.
819 * When a tree block is COW'd through a tree, there are four cases:
821 * The reference count of the block is one and the tree is the block's
822 * owner tree. Nothing to do in this case.
824 * The reference count of the block is one and the tree is not the
825 * block's owner tree. In this case, full back refs is used for pointers
826 * in the block. Remove these full back refs, add implicit back refs for
827 * every pointers in the new block.
829 * The reference count of the block is greater than one and the tree is
830 * the block's owner tree. In this case, implicit back refs is used for
831 * pointers in the block. Add full back refs for every pointers in the
832 * block, increase lower level extents' reference counts. The original
833 * implicit back refs are entailed to the new block.
835 * The reference count of the block is greater than one and the tree is
836 * not the block's owner tree. Add implicit back refs for every pointer in
837 * the new block, increase lower level extents' reference count.
839 * Back Reference Key composing:
841 * The key objectid corresponds to the first byte in the extent,
842 * The key type is used to differentiate between types of back refs.
843 * There are different meanings of the key offset for different types
846 * File extents can be referenced by:
848 * - multiple snapshots, subvolumes, or different generations in one subvol
849 * - different files inside a single subvolume
850 * - different offsets inside a file (bookend extents in file.c)
852 * The extent ref structure for the implicit back refs has fields for:
854 * - Objectid of the subvolume root
855 * - objectid of the file holding the reference
856 * - original offset in the file
857 * - how many bookend extents
859 * The key offset for the implicit back refs is hash of the first
862 * The extent ref structure for the full back refs has field for:
864 * - number of pointers in the tree leaf
866 * The key offset for the implicit back refs is the first byte of
869 * When a file extent is allocated, The implicit back refs is used.
870 * the fields are filled in:
872 * (root_key.objectid, inode objectid, offset in file, 1)
874 * When a file extent is removed file truncation, we find the
875 * corresponding implicit back refs and check the following fields:
877 * (btrfs_header_owner(leaf), inode objectid, offset in file)
879 * Btree extents can be referenced by:
881 * - Different subvolumes
883 * Both the implicit back refs and the full back refs for tree blocks
884 * only consist of key. The key offset for the implicit back refs is
885 * objectid of block's owner tree. The key offset for the full back refs
886 * is the first byte of parent block.
888 * When implicit back refs is used, information about the lowest key and
889 * level of the tree block are required. These information are stored in
890 * tree block info structure.
893 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
894 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
895 struct btrfs_root *root,
896 struct btrfs_path *path,
897 u64 owner, u32 extra_size)
899 struct btrfs_extent_item *item;
900 struct btrfs_extent_item_v0 *ei0;
901 struct btrfs_extent_ref_v0 *ref0;
902 struct btrfs_tree_block_info *bi;
903 struct extent_buffer *leaf;
904 struct btrfs_key key;
905 struct btrfs_key found_key;
906 u32 new_size = sizeof(*item);
910 leaf = path->nodes[0];
911 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
913 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
914 ei0 = btrfs_item_ptr(leaf, path->slots[0],
915 struct btrfs_extent_item_v0);
916 refs = btrfs_extent_refs_v0(leaf, ei0);
918 if (owner == (u64)-1) {
920 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
921 ret = btrfs_next_leaf(root, path);
925 leaf = path->nodes[0];
927 btrfs_item_key_to_cpu(leaf, &found_key,
929 BUG_ON(key.objectid != found_key.objectid);
930 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
934 ref0 = btrfs_item_ptr(leaf, path->slots[0],
935 struct btrfs_extent_ref_v0);
936 owner = btrfs_ref_objectid_v0(leaf, ref0);
940 btrfs_release_path(root, path);
942 if (owner < BTRFS_FIRST_FREE_OBJECTID)
943 new_size += sizeof(*bi);
945 new_size -= sizeof(*ei0);
946 ret = btrfs_search_slot(trans, root, &key, path,
947 new_size + extra_size, 1);
952 ret = btrfs_extend_item(trans, root, path, new_size);
955 leaf = path->nodes[0];
956 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
957 btrfs_set_extent_refs(leaf, item, refs);
958 /* FIXME: get real generation */
959 btrfs_set_extent_generation(leaf, item, 0);
960 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
961 btrfs_set_extent_flags(leaf, item,
962 BTRFS_EXTENT_FLAG_TREE_BLOCK |
963 BTRFS_BLOCK_FLAG_FULL_BACKREF);
964 bi = (struct btrfs_tree_block_info *)(item + 1);
965 /* FIXME: get first key of the block */
966 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
967 btrfs_set_tree_block_level(leaf, bi, (int)owner);
969 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
971 btrfs_mark_buffer_dirty(leaf);
976 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
978 u32 high_crc = ~(u32)0;
979 u32 low_crc = ~(u32)0;
982 lenum = cpu_to_le64(root_objectid);
983 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
984 lenum = cpu_to_le64(owner);
985 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
986 lenum = cpu_to_le64(offset);
987 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
989 return ((u64)high_crc << 31) ^ (u64)low_crc;
992 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
993 struct btrfs_extent_data_ref *ref)
995 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
996 btrfs_extent_data_ref_objectid(leaf, ref),
997 btrfs_extent_data_ref_offset(leaf, ref));
1000 static int match_extent_data_ref(struct extent_buffer *leaf,
1001 struct btrfs_extent_data_ref *ref,
1002 u64 root_objectid, u64 owner, u64 offset)
1004 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1005 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1006 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1011 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1012 struct btrfs_root *root,
1013 struct btrfs_path *path,
1014 u64 bytenr, u64 parent,
1016 u64 owner, u64 offset)
1018 struct btrfs_key key;
1019 struct btrfs_extent_data_ref *ref;
1020 struct extent_buffer *leaf;
1026 key.objectid = bytenr;
1028 key.type = BTRFS_SHARED_DATA_REF_KEY;
1029 key.offset = parent;
1031 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1032 key.offset = hash_extent_data_ref(root_objectid,
1037 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1046 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1047 key.type = BTRFS_EXTENT_REF_V0_KEY;
1048 btrfs_release_path(root, path);
1049 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1060 leaf = path->nodes[0];
1061 nritems = btrfs_header_nritems(leaf);
1063 if (path->slots[0] >= nritems) {
1064 ret = btrfs_next_leaf(root, path);
1070 leaf = path->nodes[0];
1071 nritems = btrfs_header_nritems(leaf);
1075 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1076 if (key.objectid != bytenr ||
1077 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1080 ref = btrfs_item_ptr(leaf, path->slots[0],
1081 struct btrfs_extent_data_ref);
1083 if (match_extent_data_ref(leaf, ref, root_objectid,
1086 btrfs_release_path(root, path);
1098 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1099 struct btrfs_root *root,
1100 struct btrfs_path *path,
1101 u64 bytenr, u64 parent,
1102 u64 root_objectid, u64 owner,
1103 u64 offset, int refs_to_add)
1105 struct btrfs_key key;
1106 struct extent_buffer *leaf;
1111 key.objectid = bytenr;
1113 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114 key.offset = parent;
1115 size = sizeof(struct btrfs_shared_data_ref);
1117 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1118 key.offset = hash_extent_data_ref(root_objectid,
1120 size = sizeof(struct btrfs_extent_data_ref);
1123 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1124 if (ret && ret != -EEXIST)
1127 leaf = path->nodes[0];
1129 struct btrfs_shared_data_ref *ref;
1130 ref = btrfs_item_ptr(leaf, path->slots[0],
1131 struct btrfs_shared_data_ref);
1133 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1135 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1136 num_refs += refs_to_add;
1137 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1140 struct btrfs_extent_data_ref *ref;
1141 while (ret == -EEXIST) {
1142 ref = btrfs_item_ptr(leaf, path->slots[0],
1143 struct btrfs_extent_data_ref);
1144 if (match_extent_data_ref(leaf, ref, root_objectid,
1147 btrfs_release_path(root, path);
1149 ret = btrfs_insert_empty_item(trans, root, path, &key,
1151 if (ret && ret != -EEXIST)
1154 leaf = path->nodes[0];
1156 ref = btrfs_item_ptr(leaf, path->slots[0],
1157 struct btrfs_extent_data_ref);
1159 btrfs_set_extent_data_ref_root(leaf, ref,
1161 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1162 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1163 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1165 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1166 num_refs += refs_to_add;
1167 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1170 btrfs_mark_buffer_dirty(leaf);
1173 btrfs_release_path(root, path);
1177 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1178 struct btrfs_root *root,
1179 struct btrfs_path *path,
1182 struct btrfs_key key;
1183 struct btrfs_extent_data_ref *ref1 = NULL;
1184 struct btrfs_shared_data_ref *ref2 = NULL;
1185 struct extent_buffer *leaf;
1189 leaf = path->nodes[0];
1190 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1192 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1193 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1194 struct btrfs_extent_data_ref);
1195 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1196 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1197 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1198 struct btrfs_shared_data_ref);
1199 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1200 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1201 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1202 struct btrfs_extent_ref_v0 *ref0;
1203 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_extent_ref_v0);
1205 num_refs = btrfs_ref_count_v0(leaf, ref0);
1211 BUG_ON(num_refs < refs_to_drop);
1212 num_refs -= refs_to_drop;
1214 if (num_refs == 0) {
1215 ret = btrfs_del_item(trans, root, path);
1217 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1218 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1219 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1220 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1221 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1223 struct btrfs_extent_ref_v0 *ref0;
1224 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1225 struct btrfs_extent_ref_v0);
1226 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1229 btrfs_mark_buffer_dirty(leaf);
1234 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1235 struct btrfs_path *path,
1236 struct btrfs_extent_inline_ref *iref)
1238 struct btrfs_key key;
1239 struct extent_buffer *leaf;
1240 struct btrfs_extent_data_ref *ref1;
1241 struct btrfs_shared_data_ref *ref2;
1244 leaf = path->nodes[0];
1245 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1247 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1248 BTRFS_EXTENT_DATA_REF_KEY) {
1249 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1250 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1252 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1253 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1255 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1256 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1257 struct btrfs_extent_data_ref);
1258 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1259 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1260 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1261 struct btrfs_shared_data_ref);
1262 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1263 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1264 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1265 struct btrfs_extent_ref_v0 *ref0;
1266 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1267 struct btrfs_extent_ref_v0);
1268 num_refs = btrfs_ref_count_v0(leaf, ref0);
1276 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1277 struct btrfs_root *root,
1278 struct btrfs_path *path,
1279 u64 bytenr, u64 parent,
1282 struct btrfs_key key;
1285 key.objectid = bytenr;
1287 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1288 key.offset = parent;
1290 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1291 key.offset = root_objectid;
1294 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1297 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1298 if (ret == -ENOENT && parent) {
1299 btrfs_release_path(root, path);
1300 key.type = BTRFS_EXTENT_REF_V0_KEY;
1301 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1309 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1310 struct btrfs_root *root,
1311 struct btrfs_path *path,
1312 u64 bytenr, u64 parent,
1315 struct btrfs_key key;
1318 key.objectid = bytenr;
1320 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1321 key.offset = parent;
1323 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1324 key.offset = root_objectid;
1327 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1328 btrfs_release_path(root, path);
1332 static inline int extent_ref_type(u64 parent, u64 owner)
1335 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1337 type = BTRFS_SHARED_BLOCK_REF_KEY;
1339 type = BTRFS_TREE_BLOCK_REF_KEY;
1342 type = BTRFS_SHARED_DATA_REF_KEY;
1344 type = BTRFS_EXTENT_DATA_REF_KEY;
1349 static int find_next_key(struct btrfs_path *path, int level,
1350 struct btrfs_key *key)
1353 for (; level < BTRFS_MAX_LEVEL; level++) {
1354 if (!path->nodes[level])
1356 if (path->slots[level] + 1 >=
1357 btrfs_header_nritems(path->nodes[level]))
1360 btrfs_item_key_to_cpu(path->nodes[level], key,
1361 path->slots[level] + 1);
1363 btrfs_node_key_to_cpu(path->nodes[level], key,
1364 path->slots[level] + 1);
1371 * look for inline back ref. if back ref is found, *ref_ret is set
1372 * to the address of inline back ref, and 0 is returned.
1374 * if back ref isn't found, *ref_ret is set to the address where it
1375 * should be inserted, and -ENOENT is returned.
1377 * if insert is true and there are too many inline back refs, the path
1378 * points to the extent item, and -EAGAIN is returned.
1380 * NOTE: inline back refs are ordered in the same way that back ref
1381 * items in the tree are ordered.
1383 static noinline_for_stack
1384 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1385 struct btrfs_root *root,
1386 struct btrfs_path *path,
1387 struct btrfs_extent_inline_ref **ref_ret,
1388 u64 bytenr, u64 num_bytes,
1389 u64 parent, u64 root_objectid,
1390 u64 owner, u64 offset, int insert)
1392 struct btrfs_key key;
1393 struct extent_buffer *leaf;
1394 struct btrfs_extent_item *ei;
1395 struct btrfs_extent_inline_ref *iref;
1406 key.objectid = bytenr;
1407 key.type = BTRFS_EXTENT_ITEM_KEY;
1408 key.offset = num_bytes;
1410 want = extent_ref_type(parent, owner);
1412 extra_size = btrfs_extent_inline_ref_size(want);
1413 path->keep_locks = 1;
1416 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1423 leaf = path->nodes[0];
1424 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1425 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1426 if (item_size < sizeof(*ei)) {
1431 ret = convert_extent_item_v0(trans, root, path, owner,
1437 leaf = path->nodes[0];
1438 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1441 BUG_ON(item_size < sizeof(*ei));
1443 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1444 flags = btrfs_extent_flags(leaf, ei);
1446 ptr = (unsigned long)(ei + 1);
1447 end = (unsigned long)ei + item_size;
1449 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1450 ptr += sizeof(struct btrfs_tree_block_info);
1453 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1462 iref = (struct btrfs_extent_inline_ref *)ptr;
1463 type = btrfs_extent_inline_ref_type(leaf, iref);
1467 ptr += btrfs_extent_inline_ref_size(type);
1471 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1472 struct btrfs_extent_data_ref *dref;
1473 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1474 if (match_extent_data_ref(leaf, dref, root_objectid,
1479 if (hash_extent_data_ref_item(leaf, dref) <
1480 hash_extent_data_ref(root_objectid, owner, offset))
1484 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1486 if (parent == ref_offset) {
1490 if (ref_offset < parent)
1493 if (root_objectid == ref_offset) {
1497 if (ref_offset < root_objectid)
1501 ptr += btrfs_extent_inline_ref_size(type);
1503 if (err == -ENOENT && insert) {
1504 if (item_size + extra_size >=
1505 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1510 * To add new inline back ref, we have to make sure
1511 * there is no corresponding back ref item.
1512 * For simplicity, we just do not add new inline back
1513 * ref if there is any kind of item for this block
1515 if (find_next_key(path, 0, &key) == 0 &&
1516 key.objectid == bytenr &&
1517 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1522 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1525 path->keep_locks = 0;
1526 btrfs_unlock_up_safe(path, 1);
1532 * helper to add new inline back ref
1534 static noinline_for_stack
1535 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1536 struct btrfs_root *root,
1537 struct btrfs_path *path,
1538 struct btrfs_extent_inline_ref *iref,
1539 u64 parent, u64 root_objectid,
1540 u64 owner, u64 offset, int refs_to_add,
1541 struct btrfs_delayed_extent_op *extent_op)
1543 struct extent_buffer *leaf;
1544 struct btrfs_extent_item *ei;
1547 unsigned long item_offset;
1553 leaf = path->nodes[0];
1554 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1555 item_offset = (unsigned long)iref - (unsigned long)ei;
1557 type = extent_ref_type(parent, owner);
1558 size = btrfs_extent_inline_ref_size(type);
1560 ret = btrfs_extend_item(trans, root, path, size);
1563 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1564 refs = btrfs_extent_refs(leaf, ei);
1565 refs += refs_to_add;
1566 btrfs_set_extent_refs(leaf, ei, refs);
1568 __run_delayed_extent_op(extent_op, leaf, ei);
1570 ptr = (unsigned long)ei + item_offset;
1571 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1572 if (ptr < end - size)
1573 memmove_extent_buffer(leaf, ptr + size, ptr,
1576 iref = (struct btrfs_extent_inline_ref *)ptr;
1577 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1578 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1579 struct btrfs_extent_data_ref *dref;
1580 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1581 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1582 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1583 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1584 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1585 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1586 struct btrfs_shared_data_ref *sref;
1587 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1588 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1589 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1590 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1591 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1593 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1595 btrfs_mark_buffer_dirty(leaf);
1599 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1600 struct btrfs_root *root,
1601 struct btrfs_path *path,
1602 struct btrfs_extent_inline_ref **ref_ret,
1603 u64 bytenr, u64 num_bytes, u64 parent,
1604 u64 root_objectid, u64 owner, u64 offset)
1608 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1609 bytenr, num_bytes, parent,
1610 root_objectid, owner, offset, 0);
1614 btrfs_release_path(root, path);
1617 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1618 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1621 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1622 root_objectid, owner, offset);
1628 * helper to update/remove inline back ref
1630 static noinline_for_stack
1631 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1632 struct btrfs_root *root,
1633 struct btrfs_path *path,
1634 struct btrfs_extent_inline_ref *iref,
1636 struct btrfs_delayed_extent_op *extent_op)
1638 struct extent_buffer *leaf;
1639 struct btrfs_extent_item *ei;
1640 struct btrfs_extent_data_ref *dref = NULL;
1641 struct btrfs_shared_data_ref *sref = NULL;
1650 leaf = path->nodes[0];
1651 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1652 refs = btrfs_extent_refs(leaf, ei);
1653 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1654 refs += refs_to_mod;
1655 btrfs_set_extent_refs(leaf, ei, refs);
1657 __run_delayed_extent_op(extent_op, leaf, ei);
1659 type = btrfs_extent_inline_ref_type(leaf, iref);
1661 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1662 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1663 refs = btrfs_extent_data_ref_count(leaf, dref);
1664 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1665 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1666 refs = btrfs_shared_data_ref_count(leaf, sref);
1669 BUG_ON(refs_to_mod != -1);
1672 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1673 refs += refs_to_mod;
1676 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1677 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1679 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1681 size = btrfs_extent_inline_ref_size(type);
1682 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1683 ptr = (unsigned long)iref;
1684 end = (unsigned long)ei + item_size;
1685 if (ptr + size < end)
1686 memmove_extent_buffer(leaf, ptr, ptr + size,
1689 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1692 btrfs_mark_buffer_dirty(leaf);
1696 static noinline_for_stack
1697 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1698 struct btrfs_root *root,
1699 struct btrfs_path *path,
1700 u64 bytenr, u64 num_bytes, u64 parent,
1701 u64 root_objectid, u64 owner,
1702 u64 offset, int refs_to_add,
1703 struct btrfs_delayed_extent_op *extent_op)
1705 struct btrfs_extent_inline_ref *iref;
1708 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1709 bytenr, num_bytes, parent,
1710 root_objectid, owner, offset, 1);
1712 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1713 ret = update_inline_extent_backref(trans, root, path, iref,
1714 refs_to_add, extent_op);
1715 } else if (ret == -ENOENT) {
1716 ret = setup_inline_extent_backref(trans, root, path, iref,
1717 parent, root_objectid,
1718 owner, offset, refs_to_add,
1724 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1725 struct btrfs_root *root,
1726 struct btrfs_path *path,
1727 u64 bytenr, u64 parent, u64 root_objectid,
1728 u64 owner, u64 offset, int refs_to_add)
1731 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1732 BUG_ON(refs_to_add != 1);
1733 ret = insert_tree_block_ref(trans, root, path, bytenr,
1734 parent, root_objectid);
1736 ret = insert_extent_data_ref(trans, root, path, bytenr,
1737 parent, root_objectid,
1738 owner, offset, refs_to_add);
1743 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1744 struct btrfs_root *root,
1745 struct btrfs_path *path,
1746 struct btrfs_extent_inline_ref *iref,
1747 int refs_to_drop, int is_data)
1751 BUG_ON(!is_data && refs_to_drop != 1);
1753 ret = update_inline_extent_backref(trans, root, path, iref,
1754 -refs_to_drop, NULL);
1755 } else if (is_data) {
1756 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1758 ret = btrfs_del_item(trans, root, path);
1763 static int btrfs_issue_discard(struct block_device *bdev,
1766 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1769 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1770 u64 num_bytes, u64 *actual_bytes)
1773 u64 discarded_bytes = 0;
1774 struct btrfs_multi_bio *multi = NULL;
1777 /* Tell the block device(s) that the sectors can be discarded */
1778 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1779 bytenr, &num_bytes, &multi, 0);
1781 struct btrfs_bio_stripe *stripe = multi->stripes;
1785 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1786 ret = btrfs_issue_discard(stripe->dev->bdev,
1790 discarded_bytes += stripe->length;
1791 else if (ret != -EOPNOTSUPP)
1796 if (discarded_bytes && ret == -EOPNOTSUPP)
1800 *actual_bytes = discarded_bytes;
1806 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1807 struct btrfs_root *root,
1808 u64 bytenr, u64 num_bytes, u64 parent,
1809 u64 root_objectid, u64 owner, u64 offset)
1812 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1813 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1815 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1816 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1817 parent, root_objectid, (int)owner,
1818 BTRFS_ADD_DELAYED_REF, NULL);
1820 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1821 parent, root_objectid, owner, offset,
1822 BTRFS_ADD_DELAYED_REF, NULL);
1827 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1828 struct btrfs_root *root,
1829 u64 bytenr, u64 num_bytes,
1830 u64 parent, u64 root_objectid,
1831 u64 owner, u64 offset, int refs_to_add,
1832 struct btrfs_delayed_extent_op *extent_op)
1834 struct btrfs_path *path;
1835 struct extent_buffer *leaf;
1836 struct btrfs_extent_item *item;
1841 path = btrfs_alloc_path();
1846 path->leave_spinning = 1;
1847 /* this will setup the path even if it fails to insert the back ref */
1848 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1849 path, bytenr, num_bytes, parent,
1850 root_objectid, owner, offset,
1851 refs_to_add, extent_op);
1855 if (ret != -EAGAIN) {
1860 leaf = path->nodes[0];
1861 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1862 refs = btrfs_extent_refs(leaf, item);
1863 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1865 __run_delayed_extent_op(extent_op, leaf, item);
1867 btrfs_mark_buffer_dirty(leaf);
1868 btrfs_release_path(root->fs_info->extent_root, path);
1871 path->leave_spinning = 1;
1873 /* now insert the actual backref */
1874 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1875 path, bytenr, parent, root_objectid,
1876 owner, offset, refs_to_add);
1879 btrfs_free_path(path);
1883 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1884 struct btrfs_root *root,
1885 struct btrfs_delayed_ref_node *node,
1886 struct btrfs_delayed_extent_op *extent_op,
1887 int insert_reserved)
1890 struct btrfs_delayed_data_ref *ref;
1891 struct btrfs_key ins;
1896 ins.objectid = node->bytenr;
1897 ins.offset = node->num_bytes;
1898 ins.type = BTRFS_EXTENT_ITEM_KEY;
1900 ref = btrfs_delayed_node_to_data_ref(node);
1901 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1902 parent = ref->parent;
1904 ref_root = ref->root;
1906 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1908 BUG_ON(extent_op->update_key);
1909 flags |= extent_op->flags_to_set;
1911 ret = alloc_reserved_file_extent(trans, root,
1912 parent, ref_root, flags,
1913 ref->objectid, ref->offset,
1914 &ins, node->ref_mod);
1915 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1916 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1917 node->num_bytes, parent,
1918 ref_root, ref->objectid,
1919 ref->offset, node->ref_mod,
1921 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1922 ret = __btrfs_free_extent(trans, root, node->bytenr,
1923 node->num_bytes, parent,
1924 ref_root, ref->objectid,
1925 ref->offset, node->ref_mod,
1933 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1934 struct extent_buffer *leaf,
1935 struct btrfs_extent_item *ei)
1937 u64 flags = btrfs_extent_flags(leaf, ei);
1938 if (extent_op->update_flags) {
1939 flags |= extent_op->flags_to_set;
1940 btrfs_set_extent_flags(leaf, ei, flags);
1943 if (extent_op->update_key) {
1944 struct btrfs_tree_block_info *bi;
1945 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1946 bi = (struct btrfs_tree_block_info *)(ei + 1);
1947 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1951 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1952 struct btrfs_root *root,
1953 struct btrfs_delayed_ref_node *node,
1954 struct btrfs_delayed_extent_op *extent_op)
1956 struct btrfs_key key;
1957 struct btrfs_path *path;
1958 struct btrfs_extent_item *ei;
1959 struct extent_buffer *leaf;
1964 path = btrfs_alloc_path();
1968 key.objectid = node->bytenr;
1969 key.type = BTRFS_EXTENT_ITEM_KEY;
1970 key.offset = node->num_bytes;
1973 path->leave_spinning = 1;
1974 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1985 leaf = path->nodes[0];
1986 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1988 if (item_size < sizeof(*ei)) {
1989 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1995 leaf = path->nodes[0];
1996 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1999 BUG_ON(item_size < sizeof(*ei));
2000 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2001 __run_delayed_extent_op(extent_op, leaf, ei);
2003 btrfs_mark_buffer_dirty(leaf);
2005 btrfs_free_path(path);
2009 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2010 struct btrfs_root *root,
2011 struct btrfs_delayed_ref_node *node,
2012 struct btrfs_delayed_extent_op *extent_op,
2013 int insert_reserved)
2016 struct btrfs_delayed_tree_ref *ref;
2017 struct btrfs_key ins;
2021 ins.objectid = node->bytenr;
2022 ins.offset = node->num_bytes;
2023 ins.type = BTRFS_EXTENT_ITEM_KEY;
2025 ref = btrfs_delayed_node_to_tree_ref(node);
2026 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2027 parent = ref->parent;
2029 ref_root = ref->root;
2031 BUG_ON(node->ref_mod != 1);
2032 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2033 BUG_ON(!extent_op || !extent_op->update_flags ||
2034 !extent_op->update_key);
2035 ret = alloc_reserved_tree_block(trans, root,
2037 extent_op->flags_to_set,
2040 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2041 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2042 node->num_bytes, parent, ref_root,
2043 ref->level, 0, 1, extent_op);
2044 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2045 ret = __btrfs_free_extent(trans, root, node->bytenr,
2046 node->num_bytes, parent, ref_root,
2047 ref->level, 0, 1, extent_op);
2054 /* helper function to actually process a single delayed ref entry */
2055 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2056 struct btrfs_root *root,
2057 struct btrfs_delayed_ref_node *node,
2058 struct btrfs_delayed_extent_op *extent_op,
2059 int insert_reserved)
2062 if (btrfs_delayed_ref_is_head(node)) {
2063 struct btrfs_delayed_ref_head *head;
2065 * we've hit the end of the chain and we were supposed
2066 * to insert this extent into the tree. But, it got
2067 * deleted before we ever needed to insert it, so all
2068 * we have to do is clean up the accounting
2071 head = btrfs_delayed_node_to_head(node);
2072 if (insert_reserved) {
2073 btrfs_pin_extent(root, node->bytenr,
2074 node->num_bytes, 1);
2075 if (head->is_data) {
2076 ret = btrfs_del_csums(trans, root,
2082 mutex_unlock(&head->mutex);
2086 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2087 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2088 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2090 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2091 node->type == BTRFS_SHARED_DATA_REF_KEY)
2092 ret = run_delayed_data_ref(trans, root, node, extent_op,
2099 static noinline struct btrfs_delayed_ref_node *
2100 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2102 struct rb_node *node;
2103 struct btrfs_delayed_ref_node *ref;
2104 int action = BTRFS_ADD_DELAYED_REF;
2107 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2108 * this prevents ref count from going down to zero when
2109 * there still are pending delayed ref.
2111 node = rb_prev(&head->node.rb_node);
2115 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2117 if (ref->bytenr != head->node.bytenr)
2119 if (ref->action == action)
2121 node = rb_prev(node);
2123 if (action == BTRFS_ADD_DELAYED_REF) {
2124 action = BTRFS_DROP_DELAYED_REF;
2130 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2131 struct btrfs_root *root,
2132 struct list_head *cluster)
2134 struct btrfs_delayed_ref_root *delayed_refs;
2135 struct btrfs_delayed_ref_node *ref;
2136 struct btrfs_delayed_ref_head *locked_ref = NULL;
2137 struct btrfs_delayed_extent_op *extent_op;
2140 int must_insert_reserved = 0;
2142 delayed_refs = &trans->transaction->delayed_refs;
2145 /* pick a new head ref from the cluster list */
2146 if (list_empty(cluster))
2149 locked_ref = list_entry(cluster->next,
2150 struct btrfs_delayed_ref_head, cluster);
2152 /* grab the lock that says we are going to process
2153 * all the refs for this head */
2154 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2157 * we may have dropped the spin lock to get the head
2158 * mutex lock, and that might have given someone else
2159 * time to free the head. If that's true, it has been
2160 * removed from our list and we can move on.
2162 if (ret == -EAGAIN) {
2170 * record the must insert reserved flag before we
2171 * drop the spin lock.
2173 must_insert_reserved = locked_ref->must_insert_reserved;
2174 locked_ref->must_insert_reserved = 0;
2176 extent_op = locked_ref->extent_op;
2177 locked_ref->extent_op = NULL;
2180 * locked_ref is the head node, so we have to go one
2181 * node back for any delayed ref updates
2183 ref = select_delayed_ref(locked_ref);
2185 /* All delayed refs have been processed, Go ahead
2186 * and send the head node to run_one_delayed_ref,
2187 * so that any accounting fixes can happen
2189 ref = &locked_ref->node;
2191 if (extent_op && must_insert_reserved) {
2197 spin_unlock(&delayed_refs->lock);
2199 ret = run_delayed_extent_op(trans, root,
2205 spin_lock(&delayed_refs->lock);
2209 list_del_init(&locked_ref->cluster);
2214 rb_erase(&ref->rb_node, &delayed_refs->root);
2215 delayed_refs->num_entries--;
2217 spin_unlock(&delayed_refs->lock);
2219 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2220 must_insert_reserved);
2223 btrfs_put_delayed_ref(ref);
2228 spin_lock(&delayed_refs->lock);
2234 * this starts processing the delayed reference count updates and
2235 * extent insertions we have queued up so far. count can be
2236 * 0, which means to process everything in the tree at the start
2237 * of the run (but not newly added entries), or it can be some target
2238 * number you'd like to process.
2240 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2241 struct btrfs_root *root, unsigned long count)
2243 struct rb_node *node;
2244 struct btrfs_delayed_ref_root *delayed_refs;
2245 struct btrfs_delayed_ref_node *ref;
2246 struct list_head cluster;
2248 int run_all = count == (unsigned long)-1;
2251 if (root == root->fs_info->extent_root)
2252 root = root->fs_info->tree_root;
2254 delayed_refs = &trans->transaction->delayed_refs;
2255 INIT_LIST_HEAD(&cluster);
2257 spin_lock(&delayed_refs->lock);
2259 count = delayed_refs->num_entries * 2;
2263 if (!(run_all || run_most) &&
2264 delayed_refs->num_heads_ready < 64)
2268 * go find something we can process in the rbtree. We start at
2269 * the beginning of the tree, and then build a cluster
2270 * of refs to process starting at the first one we are able to
2273 ret = btrfs_find_ref_cluster(trans, &cluster,
2274 delayed_refs->run_delayed_start);
2278 ret = run_clustered_refs(trans, root, &cluster);
2281 count -= min_t(unsigned long, ret, count);
2288 node = rb_first(&delayed_refs->root);
2291 count = (unsigned long)-1;
2294 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2296 if (btrfs_delayed_ref_is_head(ref)) {
2297 struct btrfs_delayed_ref_head *head;
2299 head = btrfs_delayed_node_to_head(ref);
2300 atomic_inc(&ref->refs);
2302 spin_unlock(&delayed_refs->lock);
2303 mutex_lock(&head->mutex);
2304 mutex_unlock(&head->mutex);
2306 btrfs_put_delayed_ref(ref);
2310 node = rb_next(node);
2312 spin_unlock(&delayed_refs->lock);
2313 schedule_timeout(1);
2317 spin_unlock(&delayed_refs->lock);
2321 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2322 struct btrfs_root *root,
2323 u64 bytenr, u64 num_bytes, u64 flags,
2326 struct btrfs_delayed_extent_op *extent_op;
2329 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2333 extent_op->flags_to_set = flags;
2334 extent_op->update_flags = 1;
2335 extent_op->update_key = 0;
2336 extent_op->is_data = is_data ? 1 : 0;
2338 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2344 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2345 struct btrfs_root *root,
2346 struct btrfs_path *path,
2347 u64 objectid, u64 offset, u64 bytenr)
2349 struct btrfs_delayed_ref_head *head;
2350 struct btrfs_delayed_ref_node *ref;
2351 struct btrfs_delayed_data_ref *data_ref;
2352 struct btrfs_delayed_ref_root *delayed_refs;
2353 struct rb_node *node;
2357 delayed_refs = &trans->transaction->delayed_refs;
2358 spin_lock(&delayed_refs->lock);
2359 head = btrfs_find_delayed_ref_head(trans, bytenr);
2363 if (!mutex_trylock(&head->mutex)) {
2364 atomic_inc(&head->node.refs);
2365 spin_unlock(&delayed_refs->lock);
2367 btrfs_release_path(root->fs_info->extent_root, path);
2369 mutex_lock(&head->mutex);
2370 mutex_unlock(&head->mutex);
2371 btrfs_put_delayed_ref(&head->node);
2375 node = rb_prev(&head->node.rb_node);
2379 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2381 if (ref->bytenr != bytenr)
2385 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2388 data_ref = btrfs_delayed_node_to_data_ref(ref);
2390 node = rb_prev(node);
2392 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2393 if (ref->bytenr == bytenr)
2397 if (data_ref->root != root->root_key.objectid ||
2398 data_ref->objectid != objectid || data_ref->offset != offset)
2403 mutex_unlock(&head->mutex);
2405 spin_unlock(&delayed_refs->lock);
2409 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2410 struct btrfs_root *root,
2411 struct btrfs_path *path,
2412 u64 objectid, u64 offset, u64 bytenr)
2414 struct btrfs_root *extent_root = root->fs_info->extent_root;
2415 struct extent_buffer *leaf;
2416 struct btrfs_extent_data_ref *ref;
2417 struct btrfs_extent_inline_ref *iref;
2418 struct btrfs_extent_item *ei;
2419 struct btrfs_key key;
2423 key.objectid = bytenr;
2424 key.offset = (u64)-1;
2425 key.type = BTRFS_EXTENT_ITEM_KEY;
2427 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2433 if (path->slots[0] == 0)
2437 leaf = path->nodes[0];
2438 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2440 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2444 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2445 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2446 if (item_size < sizeof(*ei)) {
2447 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2451 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2453 if (item_size != sizeof(*ei) +
2454 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2457 if (btrfs_extent_generation(leaf, ei) <=
2458 btrfs_root_last_snapshot(&root->root_item))
2461 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2462 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2463 BTRFS_EXTENT_DATA_REF_KEY)
2466 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2467 if (btrfs_extent_refs(leaf, ei) !=
2468 btrfs_extent_data_ref_count(leaf, ref) ||
2469 btrfs_extent_data_ref_root(leaf, ref) !=
2470 root->root_key.objectid ||
2471 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2472 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2480 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2481 struct btrfs_root *root,
2482 u64 objectid, u64 offset, u64 bytenr)
2484 struct btrfs_path *path;
2488 path = btrfs_alloc_path();
2493 ret = check_committed_ref(trans, root, path, objectid,
2495 if (ret && ret != -ENOENT)
2498 ret2 = check_delayed_ref(trans, root, path, objectid,
2500 } while (ret2 == -EAGAIN);
2502 if (ret2 && ret2 != -ENOENT) {
2507 if (ret != -ENOENT || ret2 != -ENOENT)
2510 btrfs_free_path(path);
2511 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2517 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2518 struct extent_buffer *buf, u32 nr_extents)
2520 struct btrfs_key key;
2521 struct btrfs_file_extent_item *fi;
2529 if (!root->ref_cows)
2532 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2534 root_gen = root->root_key.offset;
2537 root_gen = trans->transid - 1;
2540 level = btrfs_header_level(buf);
2541 nritems = btrfs_header_nritems(buf);
2544 struct btrfs_leaf_ref *ref;
2545 struct btrfs_extent_info *info;
2547 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2553 ref->root_gen = root_gen;
2554 ref->bytenr = buf->start;
2555 ref->owner = btrfs_header_owner(buf);
2556 ref->generation = btrfs_header_generation(buf);
2557 ref->nritems = nr_extents;
2558 info = ref->extents;
2560 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2562 btrfs_item_key_to_cpu(buf, &key, i);
2563 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2565 fi = btrfs_item_ptr(buf, i,
2566 struct btrfs_file_extent_item);
2567 if (btrfs_file_extent_type(buf, fi) ==
2568 BTRFS_FILE_EXTENT_INLINE)
2570 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2571 if (disk_bytenr == 0)
2574 info->bytenr = disk_bytenr;
2576 btrfs_file_extent_disk_num_bytes(buf, fi);
2577 info->objectid = key.objectid;
2578 info->offset = key.offset;
2582 ret = btrfs_add_leaf_ref(root, ref, shared);
2583 if (ret == -EEXIST && shared) {
2584 struct btrfs_leaf_ref *old;
2585 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2587 btrfs_remove_leaf_ref(root, old);
2588 btrfs_free_leaf_ref(root, old);
2589 ret = btrfs_add_leaf_ref(root, ref, shared);
2592 btrfs_free_leaf_ref(root, ref);
2598 /* when a block goes through cow, we update the reference counts of
2599 * everything that block points to. The internal pointers of the block
2600 * can be in just about any order, and it is likely to have clusters of
2601 * things that are close together and clusters of things that are not.
2603 * To help reduce the seeks that come with updating all of these reference
2604 * counts, sort them by byte number before actual updates are done.
2606 * struct refsort is used to match byte number to slot in the btree block.
2607 * we sort based on the byte number and then use the slot to actually
2610 * struct refsort is smaller than strcut btrfs_item and smaller than
2611 * struct btrfs_key_ptr. Since we're currently limited to the page size
2612 * for a btree block, there's no way for a kmalloc of refsorts for a
2613 * single node to be bigger than a page.
2621 * for passing into sort()
2623 static int refsort_cmp(const void *a_void, const void *b_void)
2625 const struct refsort *a = a_void;
2626 const struct refsort *b = b_void;
2628 if (a->bytenr < b->bytenr)
2630 if (a->bytenr > b->bytenr)
2636 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2637 struct btrfs_root *root,
2638 struct extent_buffer *buf,
2639 int full_backref, int inc)
2646 struct btrfs_key key;
2647 struct btrfs_file_extent_item *fi;
2651 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2652 u64, u64, u64, u64, u64, u64);
2654 ref_root = btrfs_header_owner(buf);
2655 nritems = btrfs_header_nritems(buf);
2656 level = btrfs_header_level(buf);
2658 if (!root->ref_cows && level == 0)
2662 process_func = btrfs_inc_extent_ref;
2664 process_func = btrfs_free_extent;
2667 parent = buf->start;
2671 for (i = 0; i < nritems; i++) {
2673 btrfs_item_key_to_cpu(buf, &key, i);
2674 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2676 fi = btrfs_item_ptr(buf, i,
2677 struct btrfs_file_extent_item);
2678 if (btrfs_file_extent_type(buf, fi) ==
2679 BTRFS_FILE_EXTENT_INLINE)
2681 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2685 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2686 key.offset -= btrfs_file_extent_offset(buf, fi);
2687 ret = process_func(trans, root, bytenr, num_bytes,
2688 parent, ref_root, key.objectid,
2693 bytenr = btrfs_node_blockptr(buf, i);
2694 num_bytes = btrfs_level_size(root, level - 1);
2695 ret = process_func(trans, root, bytenr, num_bytes,
2696 parent, ref_root, level - 1, 0);
2707 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2708 struct extent_buffer *buf, int full_backref)
2710 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2713 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2714 struct extent_buffer *buf, int full_backref)
2716 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2719 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2720 struct btrfs_root *root,
2721 struct btrfs_path *path,
2722 struct btrfs_block_group_cache *cache)
2725 struct btrfs_root *extent_root = root->fs_info->extent_root;
2727 struct extent_buffer *leaf;
2729 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2734 leaf = path->nodes[0];
2735 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2736 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2737 btrfs_mark_buffer_dirty(leaf);
2738 btrfs_release_path(extent_root, path);
2746 static struct btrfs_block_group_cache *
2747 next_block_group(struct btrfs_root *root,
2748 struct btrfs_block_group_cache *cache)
2750 struct rb_node *node;
2751 spin_lock(&root->fs_info->block_group_cache_lock);
2752 node = rb_next(&cache->cache_node);
2753 btrfs_put_block_group(cache);
2755 cache = rb_entry(node, struct btrfs_block_group_cache,
2757 btrfs_get_block_group(cache);
2760 spin_unlock(&root->fs_info->block_group_cache_lock);
2764 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2765 struct btrfs_trans_handle *trans,
2766 struct btrfs_path *path)
2768 struct btrfs_root *root = block_group->fs_info->tree_root;
2769 struct inode *inode = NULL;
2771 int dcs = BTRFS_DC_ERROR;
2777 * If this block group is smaller than 100 megs don't bother caching the
2780 if (block_group->key.offset < (100 * 1024 * 1024)) {
2781 spin_lock(&block_group->lock);
2782 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2783 spin_unlock(&block_group->lock);
2788 inode = lookup_free_space_inode(root, block_group, path);
2789 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2790 ret = PTR_ERR(inode);
2791 btrfs_release_path(root, path);
2795 if (IS_ERR(inode)) {
2799 if (block_group->ro)
2802 ret = create_free_space_inode(root, trans, block_group, path);
2809 * We want to set the generation to 0, that way if anything goes wrong
2810 * from here on out we know not to trust this cache when we load up next
2813 BTRFS_I(inode)->generation = 0;
2814 ret = btrfs_update_inode(trans, root, inode);
2817 if (i_size_read(inode) > 0) {
2818 ret = btrfs_truncate_free_space_cache(root, trans, path,
2824 spin_lock(&block_group->lock);
2825 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2826 /* We're not cached, don't bother trying to write stuff out */
2827 dcs = BTRFS_DC_WRITTEN;
2828 spin_unlock(&block_group->lock);
2831 spin_unlock(&block_group->lock);
2833 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2838 * Just to make absolutely sure we have enough space, we're going to
2839 * preallocate 12 pages worth of space for each block group. In
2840 * practice we ought to use at most 8, but we need extra space so we can
2841 * add our header and have a terminator between the extents and the
2845 num_pages *= PAGE_CACHE_SIZE;
2847 ret = btrfs_check_data_free_space(inode, num_pages);
2851 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2852 num_pages, num_pages,
2855 dcs = BTRFS_DC_SETUP;
2856 btrfs_free_reserved_data_space(inode, num_pages);
2860 btrfs_release_path(root, path);
2862 spin_lock(&block_group->lock);
2863 block_group->disk_cache_state = dcs;
2864 spin_unlock(&block_group->lock);
2869 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2870 struct btrfs_root *root)
2872 struct btrfs_block_group_cache *cache;
2874 struct btrfs_path *path;
2877 path = btrfs_alloc_path();
2883 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2885 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2887 cache = next_block_group(root, cache);
2895 err = cache_save_setup(cache, trans, path);
2896 last = cache->key.objectid + cache->key.offset;
2897 btrfs_put_block_group(cache);
2902 err = btrfs_run_delayed_refs(trans, root,
2907 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2909 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2910 btrfs_put_block_group(cache);
2916 cache = next_block_group(root, cache);
2925 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2926 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2928 last = cache->key.objectid + cache->key.offset;
2930 err = write_one_cache_group(trans, root, path, cache);
2932 btrfs_put_block_group(cache);
2937 * I don't think this is needed since we're just marking our
2938 * preallocated extent as written, but just in case it can't
2942 err = btrfs_run_delayed_refs(trans, root,
2947 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2950 * Really this shouldn't happen, but it could if we
2951 * couldn't write the entire preallocated extent and
2952 * splitting the extent resulted in a new block.
2955 btrfs_put_block_group(cache);
2958 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2960 cache = next_block_group(root, cache);
2969 btrfs_write_out_cache(root, trans, cache, path);
2972 * If we didn't have an error then the cache state is still
2973 * NEED_WRITE, so we can set it to WRITTEN.
2975 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2976 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2977 last = cache->key.objectid + cache->key.offset;
2978 btrfs_put_block_group(cache);
2981 btrfs_free_path(path);
2985 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2987 struct btrfs_block_group_cache *block_group;
2990 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2991 if (!block_group || block_group->ro)
2994 btrfs_put_block_group(block_group);
2998 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2999 u64 total_bytes, u64 bytes_used,
3000 struct btrfs_space_info **space_info)
3002 struct btrfs_space_info *found;
3006 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3007 BTRFS_BLOCK_GROUP_RAID10))
3012 found = __find_space_info(info, flags);
3014 spin_lock(&found->lock);
3015 found->total_bytes += total_bytes;
3016 found->disk_total += total_bytes * factor;
3017 found->bytes_used += bytes_used;
3018 found->disk_used += bytes_used * factor;
3020 spin_unlock(&found->lock);
3021 *space_info = found;
3024 found = kzalloc(sizeof(*found), GFP_NOFS);
3028 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3029 INIT_LIST_HEAD(&found->block_groups[i]);
3030 init_rwsem(&found->groups_sem);
3031 spin_lock_init(&found->lock);
3032 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3033 BTRFS_BLOCK_GROUP_SYSTEM |
3034 BTRFS_BLOCK_GROUP_METADATA);
3035 found->total_bytes = total_bytes;
3036 found->disk_total = total_bytes * factor;
3037 found->bytes_used = bytes_used;
3038 found->disk_used = bytes_used * factor;
3039 found->bytes_pinned = 0;
3040 found->bytes_reserved = 0;
3041 found->bytes_readonly = 0;
3042 found->bytes_may_use = 0;
3044 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3045 found->chunk_alloc = 0;
3046 *space_info = found;
3047 list_add_rcu(&found->list, &info->space_info);
3048 atomic_set(&found->caching_threads, 0);
3052 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3054 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3055 BTRFS_BLOCK_GROUP_RAID1 |
3056 BTRFS_BLOCK_GROUP_RAID10 |
3057 BTRFS_BLOCK_GROUP_DUP);
3059 if (flags & BTRFS_BLOCK_GROUP_DATA)
3060 fs_info->avail_data_alloc_bits |= extra_flags;
3061 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3062 fs_info->avail_metadata_alloc_bits |= extra_flags;
3063 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3064 fs_info->avail_system_alloc_bits |= extra_flags;
3068 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3071 * we add in the count of missing devices because we want
3072 * to make sure that any RAID levels on a degraded FS
3073 * continue to be honored.
3075 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3076 root->fs_info->fs_devices->missing_devices;
3078 if (num_devices == 1)
3079 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3080 if (num_devices < 4)
3081 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3083 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3084 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3085 BTRFS_BLOCK_GROUP_RAID10))) {
3086 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3089 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3090 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3091 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3094 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3095 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3096 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3097 (flags & BTRFS_BLOCK_GROUP_DUP)))
3098 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3102 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3104 if (flags & BTRFS_BLOCK_GROUP_DATA)
3105 flags |= root->fs_info->avail_data_alloc_bits &
3106 root->fs_info->data_alloc_profile;
3107 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3108 flags |= root->fs_info->avail_system_alloc_bits &
3109 root->fs_info->system_alloc_profile;
3110 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3111 flags |= root->fs_info->avail_metadata_alloc_bits &
3112 root->fs_info->metadata_alloc_profile;
3113 return btrfs_reduce_alloc_profile(root, flags);
3116 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3121 flags = BTRFS_BLOCK_GROUP_DATA;
3122 else if (root == root->fs_info->chunk_root)
3123 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3125 flags = BTRFS_BLOCK_GROUP_METADATA;
3127 return get_alloc_profile(root, flags);
3130 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3132 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3133 BTRFS_BLOCK_GROUP_DATA);
3137 * This will check the space that the inode allocates from to make sure we have
3138 * enough space for bytes.
3140 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3142 struct btrfs_space_info *data_sinfo;
3143 struct btrfs_root *root = BTRFS_I(inode)->root;
3145 int ret = 0, committed = 0, alloc_chunk = 1;
3147 /* make sure bytes are sectorsize aligned */
3148 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3150 if (root == root->fs_info->tree_root) {
3155 data_sinfo = BTRFS_I(inode)->space_info;
3160 /* make sure we have enough space to handle the data first */
3161 spin_lock(&data_sinfo->lock);
3162 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3163 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3164 data_sinfo->bytes_may_use;
3166 if (used + bytes > data_sinfo->total_bytes) {
3167 struct btrfs_trans_handle *trans;
3170 * if we don't have enough free bytes in this space then we need
3171 * to alloc a new chunk.
3173 if (!data_sinfo->full && alloc_chunk) {
3176 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3177 spin_unlock(&data_sinfo->lock);
3179 alloc_target = btrfs_get_alloc_profile(root, 1);
3180 trans = btrfs_join_transaction(root);
3182 return PTR_ERR(trans);
3184 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3185 bytes + 2 * 1024 * 1024,
3187 CHUNK_ALLOC_NO_FORCE);
3188 btrfs_end_transaction(trans, root);
3197 btrfs_set_inode_space_info(root, inode);
3198 data_sinfo = BTRFS_I(inode)->space_info;
3202 spin_unlock(&data_sinfo->lock);
3204 /* commit the current transaction and try again */
3207 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3209 trans = btrfs_join_transaction(root);
3211 return PTR_ERR(trans);
3212 ret = btrfs_commit_transaction(trans, root);
3218 #if 0 /* I hope we never need this code again, just in case */
3219 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3220 "%llu bytes_reserved, " "%llu bytes_pinned, "
3221 "%llu bytes_readonly, %llu may use %llu total\n",
3222 (unsigned long long)bytes,
3223 (unsigned long long)data_sinfo->bytes_used,
3224 (unsigned long long)data_sinfo->bytes_reserved,
3225 (unsigned long long)data_sinfo->bytes_pinned,
3226 (unsigned long long)data_sinfo->bytes_readonly,
3227 (unsigned long long)data_sinfo->bytes_may_use,
3228 (unsigned long long)data_sinfo->total_bytes);
3232 data_sinfo->bytes_may_use += bytes;
3233 BTRFS_I(inode)->reserved_bytes += bytes;
3234 spin_unlock(&data_sinfo->lock);
3240 * called when we are clearing an delalloc extent from the
3241 * inode's io_tree or there was an error for whatever reason
3242 * after calling btrfs_check_data_free_space
3244 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3246 struct btrfs_root *root = BTRFS_I(inode)->root;
3247 struct btrfs_space_info *data_sinfo;
3249 /* make sure bytes are sectorsize aligned */
3250 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3252 data_sinfo = BTRFS_I(inode)->space_info;
3253 spin_lock(&data_sinfo->lock);
3254 data_sinfo->bytes_may_use -= bytes;
3255 BTRFS_I(inode)->reserved_bytes -= bytes;
3256 spin_unlock(&data_sinfo->lock);
3259 static void force_metadata_allocation(struct btrfs_fs_info *info)
3261 struct list_head *head = &info->space_info;
3262 struct btrfs_space_info *found;
3265 list_for_each_entry_rcu(found, head, list) {
3266 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3267 found->force_alloc = CHUNK_ALLOC_FORCE;
3272 static int should_alloc_chunk(struct btrfs_root *root,
3273 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3276 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3277 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3280 if (force == CHUNK_ALLOC_FORCE)
3284 * in limited mode, we want to have some free space up to
3285 * about 1% of the FS size.
3287 if (force == CHUNK_ALLOC_LIMITED) {
3288 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3289 thresh = max_t(u64, 64 * 1024 * 1024,
3290 div_factor_fine(thresh, 1));
3292 if (num_bytes - num_allocated < thresh)
3297 * we have two similar checks here, one based on percentage
3298 * and once based on a hard number of 256MB. The idea
3299 * is that if we have a good amount of free
3300 * room, don't allocate a chunk. A good mount is
3301 * less than 80% utilized of the chunks we have allocated,
3302 * or more than 256MB free
3304 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3307 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3310 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3312 /* 256MB or 5% of the FS */
3313 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3315 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3320 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3321 struct btrfs_root *extent_root, u64 alloc_bytes,
3322 u64 flags, int force)
3324 struct btrfs_space_info *space_info;
3325 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3326 int wait_for_alloc = 0;
3329 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3331 space_info = __find_space_info(extent_root->fs_info, flags);
3333 ret = update_space_info(extent_root->fs_info, flags,
3337 BUG_ON(!space_info);
3340 spin_lock(&space_info->lock);
3341 if (space_info->force_alloc)
3342 force = space_info->force_alloc;
3343 if (space_info->full) {
3344 spin_unlock(&space_info->lock);
3348 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3349 spin_unlock(&space_info->lock);
3351 } else if (space_info->chunk_alloc) {
3354 space_info->chunk_alloc = 1;
3357 spin_unlock(&space_info->lock);
3359 mutex_lock(&fs_info->chunk_mutex);
3362 * The chunk_mutex is held throughout the entirety of a chunk
3363 * allocation, so once we've acquired the chunk_mutex we know that the
3364 * other guy is done and we need to recheck and see if we should
3367 if (wait_for_alloc) {
3368 mutex_unlock(&fs_info->chunk_mutex);
3374 * If we have mixed data/metadata chunks we want to make sure we keep
3375 * allocating mixed chunks instead of individual chunks.
3377 if (btrfs_mixed_space_info(space_info))
3378 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3381 * if we're doing a data chunk, go ahead and make sure that
3382 * we keep a reasonable number of metadata chunks allocated in the
3385 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3386 fs_info->data_chunk_allocations++;
3387 if (!(fs_info->data_chunk_allocations %
3388 fs_info->metadata_ratio))
3389 force_metadata_allocation(fs_info);
3392 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3393 spin_lock(&space_info->lock);
3395 space_info->full = 1;
3399 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3400 space_info->chunk_alloc = 0;
3401 spin_unlock(&space_info->lock);
3402 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3407 * shrink metadata reservation for delalloc
3409 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3410 struct btrfs_root *root, u64 to_reclaim, int sync)
3412 struct btrfs_block_rsv *block_rsv;
3413 struct btrfs_space_info *space_info;
3418 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3420 unsigned long progress;
3422 block_rsv = &root->fs_info->delalloc_block_rsv;
3423 space_info = block_rsv->space_info;
3426 reserved = space_info->bytes_reserved;
3427 progress = space_info->reservation_progress;
3432 max_reclaim = min(reserved, to_reclaim);
3434 while (loops < 1024) {
3435 /* have the flusher threads jump in and do some IO */
3437 nr_pages = min_t(unsigned long, nr_pages,
3438 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3439 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3441 spin_lock(&space_info->lock);
3442 if (reserved > space_info->bytes_reserved)
3443 reclaimed += reserved - space_info->bytes_reserved;
3444 reserved = space_info->bytes_reserved;
3445 spin_unlock(&space_info->lock);
3449 if (reserved == 0 || reclaimed >= max_reclaim)
3452 if (trans && trans->transaction->blocked)
3455 time_left = schedule_timeout_interruptible(1);
3457 /* We were interrupted, exit */
3461 /* we've kicked the IO a few times, if anything has been freed,
3462 * exit. There is no sense in looping here for a long time
3463 * when we really need to commit the transaction, or there are
3464 * just too many writers without enough free space
3469 if (progress != space_info->reservation_progress)
3474 return reclaimed >= to_reclaim;
3478 * Retries tells us how many times we've called reserve_metadata_bytes. The
3479 * idea is if this is the first call (retries == 0) then we will add to our
3480 * reserved count if we can't make the allocation in order to hold our place
3481 * while we go and try and free up space. That way for retries > 1 we don't try
3482 * and add space, we just check to see if the amount of unused space is >= the
3483 * total space, meaning that our reservation is valid.
3485 * However if we don't intend to retry this reservation, pass -1 as retries so
3486 * that it short circuits this logic.
3488 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3489 struct btrfs_root *root,
3490 struct btrfs_block_rsv *block_rsv,
3491 u64 orig_bytes, int flush)
3493 struct btrfs_space_info *space_info = block_rsv->space_info;
3495 u64 num_bytes = orig_bytes;
3498 bool reserved = false;
3499 bool committed = false;
3506 spin_lock(&space_info->lock);
3507 unused = space_info->bytes_used + space_info->bytes_reserved +
3508 space_info->bytes_pinned + space_info->bytes_readonly +
3509 space_info->bytes_may_use;
3512 * The idea here is that we've not already over-reserved the block group
3513 * then we can go ahead and save our reservation first and then start
3514 * flushing if we need to. Otherwise if we've already overcommitted
3515 * lets start flushing stuff first and then come back and try to make
3518 if (unused <= space_info->total_bytes) {
3519 unused = space_info->total_bytes - unused;
3520 if (unused >= num_bytes) {
3522 space_info->bytes_reserved += orig_bytes;
3526 * Ok set num_bytes to orig_bytes since we aren't
3527 * overocmmitted, this way we only try and reclaim what
3530 num_bytes = orig_bytes;
3534 * Ok we're over committed, set num_bytes to the overcommitted
3535 * amount plus the amount of bytes that we need for this
3538 num_bytes = unused - space_info->total_bytes +
3539 (orig_bytes * (retries + 1));
3543 * Couldn't make our reservation, save our place so while we're trying
3544 * to reclaim space we can actually use it instead of somebody else
3545 * stealing it from us.
3547 if (ret && !reserved) {
3548 space_info->bytes_reserved += orig_bytes;
3552 spin_unlock(&space_info->lock);
3561 * We do synchronous shrinking since we don't actually unreserve
3562 * metadata until after the IO is completed.
3564 ret = shrink_delalloc(trans, root, num_bytes, 1);
3571 * So if we were overcommitted it's possible that somebody else flushed
3572 * out enough space and we simply didn't have enough space to reclaim,
3573 * so go back around and try again.
3580 spin_lock(&space_info->lock);
3582 * Not enough space to be reclaimed, don't bother committing the
3585 if (space_info->bytes_pinned < orig_bytes)
3587 spin_unlock(&space_info->lock);
3592 if (trans || committed)
3596 trans = btrfs_join_transaction(root);
3599 ret = btrfs_commit_transaction(trans, root);
3608 spin_lock(&space_info->lock);
3609 space_info->bytes_reserved -= orig_bytes;
3610 spin_unlock(&space_info->lock);
3616 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3617 struct btrfs_root *root)
3619 struct btrfs_block_rsv *block_rsv;
3621 block_rsv = trans->block_rsv;
3623 block_rsv = root->block_rsv;
3626 block_rsv = &root->fs_info->empty_block_rsv;
3631 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3635 spin_lock(&block_rsv->lock);
3636 if (block_rsv->reserved >= num_bytes) {
3637 block_rsv->reserved -= num_bytes;
3638 if (block_rsv->reserved < block_rsv->size)
3639 block_rsv->full = 0;
3642 spin_unlock(&block_rsv->lock);
3646 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3647 u64 num_bytes, int update_size)
3649 spin_lock(&block_rsv->lock);
3650 block_rsv->reserved += num_bytes;
3652 block_rsv->size += num_bytes;
3653 else if (block_rsv->reserved >= block_rsv->size)
3654 block_rsv->full = 1;
3655 spin_unlock(&block_rsv->lock);
3658 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3659 struct btrfs_block_rsv *dest, u64 num_bytes)
3661 struct btrfs_space_info *space_info = block_rsv->space_info;
3663 spin_lock(&block_rsv->lock);
3664 if (num_bytes == (u64)-1)
3665 num_bytes = block_rsv->size;
3666 block_rsv->size -= num_bytes;
3667 if (block_rsv->reserved >= block_rsv->size) {
3668 num_bytes = block_rsv->reserved - block_rsv->size;
3669 block_rsv->reserved = block_rsv->size;
3670 block_rsv->full = 1;
3674 spin_unlock(&block_rsv->lock);
3676 if (num_bytes > 0) {
3678 spin_lock(&dest->lock);
3682 bytes_to_add = dest->size - dest->reserved;
3683 bytes_to_add = min(num_bytes, bytes_to_add);
3684 dest->reserved += bytes_to_add;
3685 if (dest->reserved >= dest->size)
3687 num_bytes -= bytes_to_add;
3689 spin_unlock(&dest->lock);
3692 spin_lock(&space_info->lock);
3693 space_info->bytes_reserved -= num_bytes;
3694 space_info->reservation_progress++;
3695 spin_unlock(&space_info->lock);
3700 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3701 struct btrfs_block_rsv *dst, u64 num_bytes)
3705 ret = block_rsv_use_bytes(src, num_bytes);
3709 block_rsv_add_bytes(dst, num_bytes, 1);
3713 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3715 memset(rsv, 0, sizeof(*rsv));
3716 spin_lock_init(&rsv->lock);
3717 atomic_set(&rsv->usage, 1);
3719 INIT_LIST_HEAD(&rsv->list);
3722 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3724 struct btrfs_block_rsv *block_rsv;
3725 struct btrfs_fs_info *fs_info = root->fs_info;
3727 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3731 btrfs_init_block_rsv(block_rsv);
3732 block_rsv->space_info = __find_space_info(fs_info,
3733 BTRFS_BLOCK_GROUP_METADATA);
3737 void btrfs_free_block_rsv(struct btrfs_root *root,
3738 struct btrfs_block_rsv *rsv)
3740 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3741 btrfs_block_rsv_release(root, rsv, (u64)-1);
3748 * make the block_rsv struct be able to capture freed space.
3749 * the captured space will re-add to the the block_rsv struct
3750 * after transaction commit
3752 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3753 struct btrfs_block_rsv *block_rsv)
3755 block_rsv->durable = 1;
3756 mutex_lock(&fs_info->durable_block_rsv_mutex);
3757 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3758 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3761 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3762 struct btrfs_root *root,
3763 struct btrfs_block_rsv *block_rsv,
3771 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3773 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3780 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3781 struct btrfs_root *root,
3782 struct btrfs_block_rsv *block_rsv,
3783 u64 min_reserved, int min_factor)
3786 int commit_trans = 0;
3792 spin_lock(&block_rsv->lock);
3794 num_bytes = div_factor(block_rsv->size, min_factor);
3795 if (min_reserved > num_bytes)
3796 num_bytes = min_reserved;
3798 if (block_rsv->reserved >= num_bytes) {
3801 num_bytes -= block_rsv->reserved;
3802 if (block_rsv->durable &&
3803 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3806 spin_unlock(&block_rsv->lock);
3810 if (block_rsv->refill_used) {
3811 ret = reserve_metadata_bytes(trans, root, block_rsv,
3814 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3823 trans = btrfs_join_transaction(root);
3824 BUG_ON(IS_ERR(trans));
3825 ret = btrfs_commit_transaction(trans, root);
3832 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3833 struct btrfs_block_rsv *dst_rsv,
3836 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3839 void btrfs_block_rsv_release(struct btrfs_root *root,
3840 struct btrfs_block_rsv *block_rsv,
3843 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3844 if (global_rsv->full || global_rsv == block_rsv ||
3845 block_rsv->space_info != global_rsv->space_info)
3847 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3851 * helper to calculate size of global block reservation.
3852 * the desired value is sum of space used by extent tree,
3853 * checksum tree and root tree
3855 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3857 struct btrfs_space_info *sinfo;
3861 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3864 * per tree used space accounting can be inaccuracy, so we
3867 spin_lock(&fs_info->extent_root->accounting_lock);
3868 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3869 spin_unlock(&fs_info->extent_root->accounting_lock);
3871 spin_lock(&fs_info->csum_root->accounting_lock);
3872 num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3873 spin_unlock(&fs_info->csum_root->accounting_lock);
3875 spin_lock(&fs_info->tree_root->accounting_lock);
3876 num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3877 spin_unlock(&fs_info->tree_root->accounting_lock);
3879 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3880 spin_lock(&sinfo->lock);
3881 data_used = sinfo->bytes_used;
3882 spin_unlock(&sinfo->lock);
3884 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3885 spin_lock(&sinfo->lock);
3886 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3888 meta_used = sinfo->bytes_used;
3889 spin_unlock(&sinfo->lock);
3891 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3893 num_bytes += div64_u64(data_used + meta_used, 50);
3895 if (num_bytes * 3 > meta_used)
3896 num_bytes = div64_u64(meta_used, 3);
3898 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3901 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3903 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3904 struct btrfs_space_info *sinfo = block_rsv->space_info;
3907 num_bytes = calc_global_metadata_size(fs_info);
3909 spin_lock(&block_rsv->lock);
3910 spin_lock(&sinfo->lock);
3912 block_rsv->size = num_bytes;
3914 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3915 sinfo->bytes_reserved + sinfo->bytes_readonly +
3916 sinfo->bytes_may_use;
3918 if (sinfo->total_bytes > num_bytes) {
3919 num_bytes = sinfo->total_bytes - num_bytes;
3920 block_rsv->reserved += num_bytes;
3921 sinfo->bytes_reserved += num_bytes;
3924 if (block_rsv->reserved >= block_rsv->size) {
3925 num_bytes = block_rsv->reserved - block_rsv->size;
3926 sinfo->bytes_reserved -= num_bytes;
3927 sinfo->reservation_progress++;
3928 block_rsv->reserved = block_rsv->size;
3929 block_rsv->full = 1;
3932 printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3933 block_rsv->size, block_rsv->reserved);
3935 spin_unlock(&sinfo->lock);
3936 spin_unlock(&block_rsv->lock);
3939 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3941 struct btrfs_space_info *space_info;
3943 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3944 fs_info->chunk_block_rsv.space_info = space_info;
3945 fs_info->chunk_block_rsv.priority = 10;
3947 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3948 fs_info->global_block_rsv.space_info = space_info;
3949 fs_info->global_block_rsv.priority = 10;
3950 fs_info->global_block_rsv.refill_used = 1;
3951 fs_info->delalloc_block_rsv.space_info = space_info;
3952 fs_info->trans_block_rsv.space_info = space_info;
3953 fs_info->empty_block_rsv.space_info = space_info;
3954 fs_info->empty_block_rsv.priority = 10;
3956 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3957 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3958 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3959 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3960 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3962 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3964 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3966 update_global_block_rsv(fs_info);
3969 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3971 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3972 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3973 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3974 WARN_ON(fs_info->trans_block_rsv.size > 0);
3975 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3976 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3977 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3980 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3982 return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3986 int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans,
3987 struct btrfs_root *root,
3988 struct btrfs_block_rsv *rsv)
3990 struct btrfs_block_rsv *trans_rsv = &root->fs_info->trans_block_rsv;
3995 * Truncate should be freeing data, but give us 2 items just in case it
3996 * needs to use some space. We may want to be smarter about this in the
3999 num_bytes = calc_trans_metadata_size(root, 2);
4001 /* We already have enough bytes, just return */
4002 if (rsv->reserved >= num_bytes)
4005 num_bytes -= rsv->reserved;
4008 * You should have reserved enough space before hand to do this, so this
4011 ret = block_rsv_migrate_bytes(trans_rsv, rsv, num_bytes);
4017 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
4018 struct btrfs_root *root,
4024 if (num_items == 0 || root->fs_info->chunk_root == root)
4027 num_bytes = calc_trans_metadata_size(root, num_items);
4028 ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
4031 trans->bytes_reserved += num_bytes;
4032 trans->block_rsv = &root->fs_info->trans_block_rsv;
4037 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4038 struct btrfs_root *root)
4040 if (!trans->bytes_reserved)
4043 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
4044 btrfs_block_rsv_release(root, trans->block_rsv,
4045 trans->bytes_reserved);
4046 trans->bytes_reserved = 0;
4049 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4050 struct inode *inode)
4052 struct btrfs_root *root = BTRFS_I(inode)->root;
4053 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4054 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4057 * We need to hold space in order to delete our orphan item once we've
4058 * added it, so this takes the reservation so we can release it later
4059 * when we are truly done with the orphan item.
4061 u64 num_bytes = calc_trans_metadata_size(root, 1);
4062 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4065 void btrfs_orphan_release_metadata(struct inode *inode)
4067 struct btrfs_root *root = BTRFS_I(inode)->root;
4068 u64 num_bytes = calc_trans_metadata_size(root, 1);
4069 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4072 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4073 struct btrfs_pending_snapshot *pending)
4075 struct btrfs_root *root = pending->root;
4076 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4077 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4079 * two for root back/forward refs, two for directory entries
4080 * and one for root of the snapshot.
4082 u64 num_bytes = calc_trans_metadata_size(root, 5);
4083 dst_rsv->space_info = src_rsv->space_info;
4084 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4087 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
4089 return num_bytes >>= 3;
4092 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4094 struct btrfs_root *root = BTRFS_I(inode)->root;
4095 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4098 int reserved_extents;
4101 if (btrfs_transaction_in_commit(root->fs_info))
4102 schedule_timeout(1);
4104 num_bytes = ALIGN(num_bytes, root->sectorsize);
4106 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4107 reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4109 if (nr_extents > reserved_extents) {
4110 nr_extents -= reserved_extents;
4111 to_reserve = calc_trans_metadata_size(root, nr_extents);
4117 to_reserve += calc_csum_metadata_size(inode, num_bytes);
4118 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4122 atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
4123 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4125 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4127 if (block_rsv->size > 512 * 1024 * 1024)
4128 shrink_delalloc(NULL, root, to_reserve, 0);
4133 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4135 struct btrfs_root *root = BTRFS_I(inode)->root;
4138 int reserved_extents;
4140 num_bytes = ALIGN(num_bytes, root->sectorsize);
4141 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4142 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4144 reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4148 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4149 if (nr_extents >= reserved_extents) {
4153 old = reserved_extents;
4154 nr_extents = reserved_extents - nr_extents;
4155 new = reserved_extents - nr_extents;
4156 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4157 reserved_extents, new);
4158 if (likely(old == reserved_extents))
4160 reserved_extents = old;
4163 to_free = calc_csum_metadata_size(inode, num_bytes);
4165 to_free += calc_trans_metadata_size(root, nr_extents);
4167 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4171 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4175 ret = btrfs_check_data_free_space(inode, num_bytes);
4179 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4181 btrfs_free_reserved_data_space(inode, num_bytes);
4188 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4190 btrfs_delalloc_release_metadata(inode, num_bytes);
4191 btrfs_free_reserved_data_space(inode, num_bytes);
4194 static int update_block_group(struct btrfs_trans_handle *trans,
4195 struct btrfs_root *root,
4196 u64 bytenr, u64 num_bytes, int alloc)
4198 struct btrfs_block_group_cache *cache = NULL;
4199 struct btrfs_fs_info *info = root->fs_info;
4200 u64 total = num_bytes;
4205 /* block accounting for super block */
4206 spin_lock(&info->delalloc_lock);
4207 old_val = btrfs_super_bytes_used(&info->super_copy);
4209 old_val += num_bytes;
4211 old_val -= num_bytes;
4212 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4213 spin_unlock(&info->delalloc_lock);
4216 cache = btrfs_lookup_block_group(info, bytenr);
4219 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4220 BTRFS_BLOCK_GROUP_RAID1 |
4221 BTRFS_BLOCK_GROUP_RAID10))
4226 * If this block group has free space cache written out, we
4227 * need to make sure to load it if we are removing space. This
4228 * is because we need the unpinning stage to actually add the
4229 * space back to the block group, otherwise we will leak space.
4231 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4232 cache_block_group(cache, trans, NULL, 1);
4234 byte_in_group = bytenr - cache->key.objectid;
4235 WARN_ON(byte_in_group > cache->key.offset);
4237 spin_lock(&cache->space_info->lock);
4238 spin_lock(&cache->lock);
4240 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4241 cache->disk_cache_state < BTRFS_DC_CLEAR)
4242 cache->disk_cache_state = BTRFS_DC_CLEAR;
4245 old_val = btrfs_block_group_used(&cache->item);
4246 num_bytes = min(total, cache->key.offset - byte_in_group);
4248 old_val += num_bytes;
4249 btrfs_set_block_group_used(&cache->item, old_val);
4250 cache->reserved -= num_bytes;
4251 cache->space_info->bytes_reserved -= num_bytes;
4252 cache->space_info->reservation_progress++;
4253 cache->space_info->bytes_used += num_bytes;
4254 cache->space_info->disk_used += num_bytes * factor;
4255 spin_unlock(&cache->lock);
4256 spin_unlock(&cache->space_info->lock);
4258 old_val -= num_bytes;
4259 btrfs_set_block_group_used(&cache->item, old_val);
4260 cache->pinned += num_bytes;
4261 cache->space_info->bytes_pinned += num_bytes;
4262 cache->space_info->bytes_used -= num_bytes;
4263 cache->space_info->disk_used -= num_bytes * factor;
4264 spin_unlock(&cache->lock);
4265 spin_unlock(&cache->space_info->lock);
4267 set_extent_dirty(info->pinned_extents,
4268 bytenr, bytenr + num_bytes - 1,
4269 GFP_NOFS | __GFP_NOFAIL);
4271 btrfs_put_block_group(cache);
4273 bytenr += num_bytes;
4278 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4280 struct btrfs_block_group_cache *cache;
4283 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4287 bytenr = cache->key.objectid;
4288 btrfs_put_block_group(cache);
4293 static int pin_down_extent(struct btrfs_root *root,
4294 struct btrfs_block_group_cache *cache,
4295 u64 bytenr, u64 num_bytes, int reserved)
4297 spin_lock(&cache->space_info->lock);
4298 spin_lock(&cache->lock);
4299 cache->pinned += num_bytes;
4300 cache->space_info->bytes_pinned += num_bytes;
4302 cache->reserved -= num_bytes;
4303 cache->space_info->bytes_reserved -= num_bytes;
4304 cache->space_info->reservation_progress++;
4306 spin_unlock(&cache->lock);
4307 spin_unlock(&cache->space_info->lock);
4309 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4310 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4315 * this function must be called within transaction
4317 int btrfs_pin_extent(struct btrfs_root *root,
4318 u64 bytenr, u64 num_bytes, int reserved)
4320 struct btrfs_block_group_cache *cache;
4322 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4325 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4327 btrfs_put_block_group(cache);
4332 * update size of reserved extents. this function may return -EAGAIN
4333 * if 'reserve' is true or 'sinfo' is false.
4335 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4336 u64 num_bytes, int reserve, int sinfo)
4340 struct btrfs_space_info *space_info = cache->space_info;
4341 spin_lock(&space_info->lock);
4342 spin_lock(&cache->lock);
4347 cache->reserved += num_bytes;
4348 space_info->bytes_reserved += num_bytes;
4352 space_info->bytes_readonly += num_bytes;
4353 cache->reserved -= num_bytes;
4354 space_info->bytes_reserved -= num_bytes;
4355 space_info->reservation_progress++;
4357 spin_unlock(&cache->lock);
4358 spin_unlock(&space_info->lock);
4360 spin_lock(&cache->lock);
4365 cache->reserved += num_bytes;
4367 cache->reserved -= num_bytes;
4369 spin_unlock(&cache->lock);
4374 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4375 struct btrfs_root *root)
4377 struct btrfs_fs_info *fs_info = root->fs_info;
4378 struct btrfs_caching_control *next;
4379 struct btrfs_caching_control *caching_ctl;
4380 struct btrfs_block_group_cache *cache;
4382 down_write(&fs_info->extent_commit_sem);
4384 list_for_each_entry_safe(caching_ctl, next,
4385 &fs_info->caching_block_groups, list) {
4386 cache = caching_ctl->block_group;
4387 if (block_group_cache_done(cache)) {
4388 cache->last_byte_to_unpin = (u64)-1;
4389 list_del_init(&caching_ctl->list);
4390 put_caching_control(caching_ctl);
4392 cache->last_byte_to_unpin = caching_ctl->progress;
4396 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4397 fs_info->pinned_extents = &fs_info->freed_extents[1];
4399 fs_info->pinned_extents = &fs_info->freed_extents[0];
4401 up_write(&fs_info->extent_commit_sem);
4403 update_global_block_rsv(fs_info);
4407 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4409 struct btrfs_fs_info *fs_info = root->fs_info;
4410 struct btrfs_block_group_cache *cache = NULL;
4413 while (start <= end) {
4415 start >= cache->key.objectid + cache->key.offset) {
4417 btrfs_put_block_group(cache);
4418 cache = btrfs_lookup_block_group(fs_info, start);
4422 len = cache->key.objectid + cache->key.offset - start;
4423 len = min(len, end + 1 - start);
4425 if (start < cache->last_byte_to_unpin) {
4426 len = min(len, cache->last_byte_to_unpin - start);
4427 btrfs_add_free_space(cache, start, len);
4432 spin_lock(&cache->space_info->lock);
4433 spin_lock(&cache->lock);
4434 cache->pinned -= len;
4435 cache->space_info->bytes_pinned -= len;
4437 cache->space_info->bytes_readonly += len;
4438 } else if (cache->reserved_pinned > 0) {
4439 len = min(len, cache->reserved_pinned);
4440 cache->reserved_pinned -= len;
4441 cache->space_info->bytes_reserved += len;
4443 spin_unlock(&cache->lock);
4444 spin_unlock(&cache->space_info->lock);
4448 btrfs_put_block_group(cache);
4452 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4453 struct btrfs_root *root)
4455 struct btrfs_fs_info *fs_info = root->fs_info;
4456 struct extent_io_tree *unpin;
4457 struct btrfs_block_rsv *block_rsv;
4458 struct btrfs_block_rsv *next_rsv;
4464 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4465 unpin = &fs_info->freed_extents[1];
4467 unpin = &fs_info->freed_extents[0];
4470 ret = find_first_extent_bit(unpin, 0, &start, &end,
4475 if (btrfs_test_opt(root, DISCARD))
4476 ret = btrfs_discard_extent(root, start,
4477 end + 1 - start, NULL);
4479 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4480 unpin_extent_range(root, start, end);
4484 mutex_lock(&fs_info->durable_block_rsv_mutex);
4485 list_for_each_entry_safe(block_rsv, next_rsv,
4486 &fs_info->durable_block_rsv_list, list) {
4488 idx = trans->transid & 0x1;
4489 if (block_rsv->freed[idx] > 0) {
4490 block_rsv_add_bytes(block_rsv,
4491 block_rsv->freed[idx], 0);
4492 block_rsv->freed[idx] = 0;
4494 if (atomic_read(&block_rsv->usage) == 0) {
4495 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4497 if (block_rsv->freed[0] == 0 &&
4498 block_rsv->freed[1] == 0) {
4499 list_del_init(&block_rsv->list);
4503 btrfs_block_rsv_release(root, block_rsv, 0);
4506 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4511 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4512 struct btrfs_root *root,
4513 u64 bytenr, u64 num_bytes, u64 parent,
4514 u64 root_objectid, u64 owner_objectid,
4515 u64 owner_offset, int refs_to_drop,
4516 struct btrfs_delayed_extent_op *extent_op)
4518 struct btrfs_key key;
4519 struct btrfs_path *path;
4520 struct btrfs_fs_info *info = root->fs_info;
4521 struct btrfs_root *extent_root = info->extent_root;
4522 struct extent_buffer *leaf;
4523 struct btrfs_extent_item *ei;
4524 struct btrfs_extent_inline_ref *iref;
4527 int extent_slot = 0;
4528 int found_extent = 0;
4533 path = btrfs_alloc_path();
4538 path->leave_spinning = 1;
4540 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4541 BUG_ON(!is_data && refs_to_drop != 1);
4543 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4544 bytenr, num_bytes, parent,
4545 root_objectid, owner_objectid,
4548 extent_slot = path->slots[0];
4549 while (extent_slot >= 0) {
4550 btrfs_item_key_to_cpu(path->nodes[0], &key,
4552 if (key.objectid != bytenr)
4554 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4555 key.offset == num_bytes) {
4559 if (path->slots[0] - extent_slot > 5)
4563 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4564 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4565 if (found_extent && item_size < sizeof(*ei))
4568 if (!found_extent) {
4570 ret = remove_extent_backref(trans, extent_root, path,
4574 btrfs_release_path(extent_root, path);
4575 path->leave_spinning = 1;
4577 key.objectid = bytenr;
4578 key.type = BTRFS_EXTENT_ITEM_KEY;
4579 key.offset = num_bytes;
4581 ret = btrfs_search_slot(trans, extent_root,
4584 printk(KERN_ERR "umm, got %d back from search"
4585 ", was looking for %llu\n", ret,
4586 (unsigned long long)bytenr);
4587 btrfs_print_leaf(extent_root, path->nodes[0]);
4590 extent_slot = path->slots[0];
4593 btrfs_print_leaf(extent_root, path->nodes[0]);
4595 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4596 "parent %llu root %llu owner %llu offset %llu\n",
4597 (unsigned long long)bytenr,
4598 (unsigned long long)parent,
4599 (unsigned long long)root_objectid,
4600 (unsigned long long)owner_objectid,
4601 (unsigned long long)owner_offset);
4604 leaf = path->nodes[0];
4605 item_size = btrfs_item_size_nr(leaf, extent_slot);
4606 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4607 if (item_size < sizeof(*ei)) {
4608 BUG_ON(found_extent || extent_slot != path->slots[0]);
4609 ret = convert_extent_item_v0(trans, extent_root, path,
4613 btrfs_release_path(extent_root, path);
4614 path->leave_spinning = 1;
4616 key.objectid = bytenr;
4617 key.type = BTRFS_EXTENT_ITEM_KEY;
4618 key.offset = num_bytes;
4620 ret = btrfs_search_slot(trans, extent_root, &key, path,
4623 printk(KERN_ERR "umm, got %d back from search"
4624 ", was looking for %llu\n", ret,
4625 (unsigned long long)bytenr);
4626 btrfs_print_leaf(extent_root, path->nodes[0]);
4629 extent_slot = path->slots[0];
4630 leaf = path->nodes[0];
4631 item_size = btrfs_item_size_nr(leaf, extent_slot);
4634 BUG_ON(item_size < sizeof(*ei));
4635 ei = btrfs_item_ptr(leaf, extent_slot,
4636 struct btrfs_extent_item);
4637 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4638 struct btrfs_tree_block_info *bi;
4639 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4640 bi = (struct btrfs_tree_block_info *)(ei + 1);
4641 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4644 refs = btrfs_extent_refs(leaf, ei);
4645 BUG_ON(refs < refs_to_drop);
4646 refs -= refs_to_drop;
4650 __run_delayed_extent_op(extent_op, leaf, ei);
4652 * In the case of inline back ref, reference count will
4653 * be updated by remove_extent_backref
4656 BUG_ON(!found_extent);
4658 btrfs_set_extent_refs(leaf, ei, refs);
4659 btrfs_mark_buffer_dirty(leaf);
4662 ret = remove_extent_backref(trans, extent_root, path,
4669 BUG_ON(is_data && refs_to_drop !=
4670 extent_data_ref_count(root, path, iref));
4672 BUG_ON(path->slots[0] != extent_slot);
4674 BUG_ON(path->slots[0] != extent_slot + 1);
4675 path->slots[0] = extent_slot;
4680 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4683 btrfs_release_path(extent_root, path);
4686 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4689 invalidate_mapping_pages(info->btree_inode->i_mapping,
4690 bytenr >> PAGE_CACHE_SHIFT,
4691 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4694 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4697 btrfs_free_path(path);
4702 * when we free an block, it is possible (and likely) that we free the last
4703 * delayed ref for that extent as well. This searches the delayed ref tree for
4704 * a given extent, and if there are no other delayed refs to be processed, it
4705 * removes it from the tree.
4707 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4708 struct btrfs_root *root, u64 bytenr)
4710 struct btrfs_delayed_ref_head *head;
4711 struct btrfs_delayed_ref_root *delayed_refs;
4712 struct btrfs_delayed_ref_node *ref;
4713 struct rb_node *node;
4716 delayed_refs = &trans->transaction->delayed_refs;
4717 spin_lock(&delayed_refs->lock);
4718 head = btrfs_find_delayed_ref_head(trans, bytenr);
4722 node = rb_prev(&head->node.rb_node);
4726 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4728 /* there are still entries for this ref, we can't drop it */
4729 if (ref->bytenr == bytenr)
4732 if (head->extent_op) {
4733 if (!head->must_insert_reserved)
4735 kfree(head->extent_op);
4736 head->extent_op = NULL;
4740 * waiting for the lock here would deadlock. If someone else has it
4741 * locked they are already in the process of dropping it anyway
4743 if (!mutex_trylock(&head->mutex))
4747 * at this point we have a head with no other entries. Go
4748 * ahead and process it.
4750 head->node.in_tree = 0;
4751 rb_erase(&head->node.rb_node, &delayed_refs->root);
4753 delayed_refs->num_entries--;
4756 * we don't take a ref on the node because we're removing it from the
4757 * tree, so we just steal the ref the tree was holding.
4759 delayed_refs->num_heads--;
4760 if (list_empty(&head->cluster))
4761 delayed_refs->num_heads_ready--;
4763 list_del_init(&head->cluster);
4764 spin_unlock(&delayed_refs->lock);
4766 BUG_ON(head->extent_op);
4767 if (head->must_insert_reserved)
4770 mutex_unlock(&head->mutex);
4771 btrfs_put_delayed_ref(&head->node);
4774 spin_unlock(&delayed_refs->lock);
4778 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4779 struct btrfs_root *root,
4780 struct extent_buffer *buf,
4781 u64 parent, int last_ref)
4783 struct btrfs_block_rsv *block_rsv;
4784 struct btrfs_block_group_cache *cache = NULL;
4787 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4788 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4789 parent, root->root_key.objectid,
4790 btrfs_header_level(buf),
4791 BTRFS_DROP_DELAYED_REF, NULL);
4798 block_rsv = get_block_rsv(trans, root);
4799 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4800 if (block_rsv->space_info != cache->space_info)
4803 if (btrfs_header_generation(buf) == trans->transid) {
4804 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4805 ret = check_ref_cleanup(trans, root, buf->start);
4810 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4811 pin_down_extent(root, cache, buf->start, buf->len, 1);
4815 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4817 btrfs_add_free_space(cache, buf->start, buf->len);
4818 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4819 if (ret == -EAGAIN) {
4820 /* block group became read-only */
4821 btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4826 spin_lock(&block_rsv->lock);
4827 if (block_rsv->reserved < block_rsv->size) {
4828 block_rsv->reserved += buf->len;
4831 spin_unlock(&block_rsv->lock);
4834 spin_lock(&cache->space_info->lock);
4835 cache->space_info->bytes_reserved -= buf->len;
4836 cache->space_info->reservation_progress++;
4837 spin_unlock(&cache->space_info->lock);
4842 if (block_rsv->durable && !cache->ro) {
4844 spin_lock(&cache->lock);
4846 cache->reserved_pinned += buf->len;
4849 spin_unlock(&cache->lock);
4852 spin_lock(&block_rsv->lock);
4853 block_rsv->freed[trans->transid & 0x1] += buf->len;
4854 spin_unlock(&block_rsv->lock);
4859 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4862 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4863 btrfs_put_block_group(cache);
4866 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4867 struct btrfs_root *root,
4868 u64 bytenr, u64 num_bytes, u64 parent,
4869 u64 root_objectid, u64 owner, u64 offset)
4874 * tree log blocks never actually go into the extent allocation
4875 * tree, just update pinning info and exit early.
4877 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4878 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4879 /* unlocks the pinned mutex */
4880 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4882 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4883 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4884 parent, root_objectid, (int)owner,
4885 BTRFS_DROP_DELAYED_REF, NULL);
4888 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4889 parent, root_objectid, owner,
4890 offset, BTRFS_DROP_DELAYED_REF, NULL);
4896 static u64 stripe_align(struct btrfs_root *root, u64 val)
4898 u64 mask = ((u64)root->stripesize - 1);
4899 u64 ret = (val + mask) & ~mask;
4904 * when we wait for progress in the block group caching, its because
4905 * our allocation attempt failed at least once. So, we must sleep
4906 * and let some progress happen before we try again.
4908 * This function will sleep at least once waiting for new free space to
4909 * show up, and then it will check the block group free space numbers
4910 * for our min num_bytes. Another option is to have it go ahead
4911 * and look in the rbtree for a free extent of a given size, but this
4915 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4918 struct btrfs_caching_control *caching_ctl;
4921 caching_ctl = get_caching_control(cache);
4925 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4926 (cache->free_space >= num_bytes));
4928 put_caching_control(caching_ctl);
4933 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4935 struct btrfs_caching_control *caching_ctl;
4938 caching_ctl = get_caching_control(cache);
4942 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4944 put_caching_control(caching_ctl);
4948 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4951 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4953 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4955 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4957 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4964 enum btrfs_loop_type {
4965 LOOP_FIND_IDEAL = 0,
4966 LOOP_CACHING_NOWAIT = 1,
4967 LOOP_CACHING_WAIT = 2,
4968 LOOP_ALLOC_CHUNK = 3,
4969 LOOP_NO_EMPTY_SIZE = 4,
4973 * walks the btree of allocated extents and find a hole of a given size.
4974 * The key ins is changed to record the hole:
4975 * ins->objectid == block start
4976 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4977 * ins->offset == number of blocks
4978 * Any available blocks before search_start are skipped.
4980 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4981 struct btrfs_root *orig_root,
4982 u64 num_bytes, u64 empty_size,
4983 u64 search_start, u64 search_end,
4984 u64 hint_byte, struct btrfs_key *ins,
4988 struct btrfs_root *root = orig_root->fs_info->extent_root;
4989 struct btrfs_free_cluster *last_ptr = NULL;
4990 struct btrfs_block_group_cache *block_group = NULL;
4991 int empty_cluster = 2 * 1024 * 1024;
4992 int allowed_chunk_alloc = 0;
4993 int done_chunk_alloc = 0;
4994 struct btrfs_space_info *space_info;
4995 int last_ptr_loop = 0;
4998 bool found_uncached_bg = false;
4999 bool failed_cluster_refill = false;
5000 bool failed_alloc = false;
5001 bool use_cluster = true;
5002 u64 ideal_cache_percent = 0;
5003 u64 ideal_cache_offset = 0;
5005 WARN_ON(num_bytes < root->sectorsize);
5006 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5010 space_info = __find_space_info(root->fs_info, data);
5012 printk(KERN_ERR "No space info for %d\n", data);
5017 * If the space info is for both data and metadata it means we have a
5018 * small filesystem and we can't use the clustering stuff.
5020 if (btrfs_mixed_space_info(space_info))
5021 use_cluster = false;
5023 if (orig_root->ref_cows || empty_size)
5024 allowed_chunk_alloc = 1;
5026 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5027 last_ptr = &root->fs_info->meta_alloc_cluster;
5028 if (!btrfs_test_opt(root, SSD))
5029 empty_cluster = 64 * 1024;
5032 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5033 btrfs_test_opt(root, SSD)) {
5034 last_ptr = &root->fs_info->data_alloc_cluster;
5038 spin_lock(&last_ptr->lock);
5039 if (last_ptr->block_group)
5040 hint_byte = last_ptr->window_start;
5041 spin_unlock(&last_ptr->lock);
5044 search_start = max(search_start, first_logical_byte(root, 0));
5045 search_start = max(search_start, hint_byte);
5050 if (search_start == hint_byte) {
5052 block_group = btrfs_lookup_block_group(root->fs_info,
5055 * we don't want to use the block group if it doesn't match our
5056 * allocation bits, or if its not cached.
5058 * However if we are re-searching with an ideal block group
5059 * picked out then we don't care that the block group is cached.
5061 if (block_group && block_group_bits(block_group, data) &&
5062 (block_group->cached != BTRFS_CACHE_NO ||
5063 search_start == ideal_cache_offset)) {
5064 down_read(&space_info->groups_sem);
5065 if (list_empty(&block_group->list) ||
5068 * someone is removing this block group,
5069 * we can't jump into the have_block_group
5070 * target because our list pointers are not
5073 btrfs_put_block_group(block_group);
5074 up_read(&space_info->groups_sem);
5076 index = get_block_group_index(block_group);
5077 goto have_block_group;
5079 } else if (block_group) {
5080 btrfs_put_block_group(block_group);
5084 down_read(&space_info->groups_sem);
5085 list_for_each_entry(block_group, &space_info->block_groups[index],
5090 btrfs_get_block_group(block_group);
5091 search_start = block_group->key.objectid;
5094 * this can happen if we end up cycling through all the
5095 * raid types, but we want to make sure we only allocate
5096 * for the proper type.
5098 if (!block_group_bits(block_group, data)) {
5099 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5100 BTRFS_BLOCK_GROUP_RAID1 |
5101 BTRFS_BLOCK_GROUP_RAID10;
5104 * if they asked for extra copies and this block group
5105 * doesn't provide them, bail. This does allow us to
5106 * fill raid0 from raid1.
5108 if ((data & extra) && !(block_group->flags & extra))
5113 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5116 ret = cache_block_group(block_group, trans,
5118 if (block_group->cached == BTRFS_CACHE_FINISHED)
5119 goto have_block_group;
5121 free_percent = btrfs_block_group_used(&block_group->item);
5122 free_percent *= 100;
5123 free_percent = div64_u64(free_percent,
5124 block_group->key.offset);
5125 free_percent = 100 - free_percent;
5126 if (free_percent > ideal_cache_percent &&
5127 likely(!block_group->ro)) {
5128 ideal_cache_offset = block_group->key.objectid;
5129 ideal_cache_percent = free_percent;
5133 * We only want to start kthread caching if we are at
5134 * the point where we will wait for caching to make
5135 * progress, or if our ideal search is over and we've
5136 * found somebody to start caching.
5138 if (loop > LOOP_CACHING_NOWAIT ||
5139 (loop > LOOP_FIND_IDEAL &&
5140 atomic_read(&space_info->caching_threads) < 2)) {
5141 ret = cache_block_group(block_group, trans,
5145 found_uncached_bg = true;
5148 * If loop is set for cached only, try the next block
5151 if (loop == LOOP_FIND_IDEAL)
5155 cached = block_group_cache_done(block_group);
5156 if (unlikely(!cached))
5157 found_uncached_bg = true;
5159 if (unlikely(block_group->ro))
5162 spin_lock(&block_group->tree_lock);
5164 block_group->free_space < num_bytes + empty_size) {
5165 spin_unlock(&block_group->tree_lock);
5168 spin_unlock(&block_group->tree_lock);
5171 * Ok we want to try and use the cluster allocator, so lets look
5172 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5173 * have tried the cluster allocator plenty of times at this
5174 * point and not have found anything, so we are likely way too
5175 * fragmented for the clustering stuff to find anything, so lets
5176 * just skip it and let the allocator find whatever block it can
5179 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5181 * the refill lock keeps out other
5182 * people trying to start a new cluster
5184 spin_lock(&last_ptr->refill_lock);
5185 if (last_ptr->block_group &&
5186 (last_ptr->block_group->ro ||
5187 !block_group_bits(last_ptr->block_group, data))) {
5189 goto refill_cluster;
5192 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5193 num_bytes, search_start);
5195 /* we have a block, we're done */
5196 spin_unlock(&last_ptr->refill_lock);
5200 spin_lock(&last_ptr->lock);
5202 * whoops, this cluster doesn't actually point to
5203 * this block group. Get a ref on the block
5204 * group is does point to and try again
5206 if (!last_ptr_loop && last_ptr->block_group &&
5207 last_ptr->block_group != block_group) {
5209 btrfs_put_block_group(block_group);
5210 block_group = last_ptr->block_group;
5211 btrfs_get_block_group(block_group);
5212 spin_unlock(&last_ptr->lock);
5213 spin_unlock(&last_ptr->refill_lock);
5216 search_start = block_group->key.objectid;
5218 * we know this block group is properly
5219 * in the list because
5220 * btrfs_remove_block_group, drops the
5221 * cluster before it removes the block
5222 * group from the list
5224 goto have_block_group;
5226 spin_unlock(&last_ptr->lock);
5229 * this cluster didn't work out, free it and
5232 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5236 /* allocate a cluster in this block group */
5237 ret = btrfs_find_space_cluster(trans, root,
5238 block_group, last_ptr,
5240 empty_cluster + empty_size);
5243 * now pull our allocation out of this
5246 offset = btrfs_alloc_from_cluster(block_group,
5247 last_ptr, num_bytes,
5250 /* we found one, proceed */
5251 spin_unlock(&last_ptr->refill_lock);
5254 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5255 && !failed_cluster_refill) {
5256 spin_unlock(&last_ptr->refill_lock);
5258 failed_cluster_refill = true;
5259 wait_block_group_cache_progress(block_group,
5260 num_bytes + empty_cluster + empty_size);
5261 goto have_block_group;
5265 * at this point we either didn't find a cluster
5266 * or we weren't able to allocate a block from our
5267 * cluster. Free the cluster we've been trying
5268 * to use, and go to the next block group
5270 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5271 spin_unlock(&last_ptr->refill_lock);
5275 offset = btrfs_find_space_for_alloc(block_group, search_start,
5276 num_bytes, empty_size);
5278 * If we didn't find a chunk, and we haven't failed on this
5279 * block group before, and this block group is in the middle of
5280 * caching and we are ok with waiting, then go ahead and wait
5281 * for progress to be made, and set failed_alloc to true.
5283 * If failed_alloc is true then we've already waited on this
5284 * block group once and should move on to the next block group.
5286 if (!offset && !failed_alloc && !cached &&
5287 loop > LOOP_CACHING_NOWAIT) {
5288 wait_block_group_cache_progress(block_group,
5289 num_bytes + empty_size);
5290 failed_alloc = true;
5291 goto have_block_group;
5292 } else if (!offset) {
5296 search_start = stripe_align(root, offset);
5297 /* move on to the next group */
5298 if (search_start + num_bytes >= search_end) {
5299 btrfs_add_free_space(block_group, offset, num_bytes);
5303 /* move on to the next group */
5304 if (search_start + num_bytes >
5305 block_group->key.objectid + block_group->key.offset) {
5306 btrfs_add_free_space(block_group, offset, num_bytes);
5310 ins->objectid = search_start;
5311 ins->offset = num_bytes;
5313 if (offset < search_start)
5314 btrfs_add_free_space(block_group, offset,
5315 search_start - offset);
5316 BUG_ON(offset > search_start);
5318 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5319 (data & BTRFS_BLOCK_GROUP_DATA));
5320 if (ret == -EAGAIN) {
5321 btrfs_add_free_space(block_group, offset, num_bytes);
5325 /* we are all good, lets return */
5326 ins->objectid = search_start;
5327 ins->offset = num_bytes;
5329 if (offset < search_start)
5330 btrfs_add_free_space(block_group, offset,
5331 search_start - offset);
5332 BUG_ON(offset > search_start);
5333 btrfs_put_block_group(block_group);
5336 failed_cluster_refill = false;
5337 failed_alloc = false;
5338 BUG_ON(index != get_block_group_index(block_group));
5339 btrfs_put_block_group(block_group);
5341 up_read(&space_info->groups_sem);
5343 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5346 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5347 * for them to make caching progress. Also
5348 * determine the best possible bg to cache
5349 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5350 * caching kthreads as we move along
5351 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5352 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5353 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5356 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5357 (found_uncached_bg || empty_size || empty_cluster ||
5358 allowed_chunk_alloc)) {
5360 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5361 found_uncached_bg = false;
5363 if (!ideal_cache_percent &&
5364 atomic_read(&space_info->caching_threads))
5368 * 1 of the following 2 things have happened so far
5370 * 1) We found an ideal block group for caching that
5371 * is mostly full and will cache quickly, so we might
5372 * as well wait for it.
5374 * 2) We searched for cached only and we didn't find
5375 * anything, and we didn't start any caching kthreads
5376 * either, so chances are we will loop through and
5377 * start a couple caching kthreads, and then come back
5378 * around and just wait for them. This will be slower
5379 * because we will have 2 caching kthreads reading at
5380 * the same time when we could have just started one
5381 * and waited for it to get far enough to give us an
5382 * allocation, so go ahead and go to the wait caching
5385 loop = LOOP_CACHING_WAIT;
5386 search_start = ideal_cache_offset;
5387 ideal_cache_percent = 0;
5389 } else if (loop == LOOP_FIND_IDEAL) {
5391 * Didn't find a uncached bg, wait on anything we find
5394 loop = LOOP_CACHING_WAIT;
5398 if (loop < LOOP_CACHING_WAIT) {
5403 if (loop == LOOP_ALLOC_CHUNK) {
5408 if (allowed_chunk_alloc) {
5409 ret = do_chunk_alloc(trans, root, num_bytes +
5410 2 * 1024 * 1024, data,
5411 CHUNK_ALLOC_LIMITED);
5412 allowed_chunk_alloc = 0;
5413 done_chunk_alloc = 1;
5414 } else if (!done_chunk_alloc &&
5415 space_info->force_alloc == CHUNK_ALLOC_NO_FORCE) {
5416 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5419 if (loop < LOOP_NO_EMPTY_SIZE) {
5424 } else if (!ins->objectid) {
5426 } else if (ins->objectid) {
5433 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5434 int dump_block_groups)
5436 struct btrfs_block_group_cache *cache;
5439 spin_lock(&info->lock);
5440 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5441 (unsigned long long)(info->total_bytes - info->bytes_used -
5442 info->bytes_pinned - info->bytes_reserved -
5443 info->bytes_readonly),
5444 (info->full) ? "" : "not ");
5445 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5446 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5447 (unsigned long long)info->total_bytes,
5448 (unsigned long long)info->bytes_used,
5449 (unsigned long long)info->bytes_pinned,
5450 (unsigned long long)info->bytes_reserved,
5451 (unsigned long long)info->bytes_may_use,
5452 (unsigned long long)info->bytes_readonly);
5453 spin_unlock(&info->lock);
5455 if (!dump_block_groups)
5458 down_read(&info->groups_sem);
5460 list_for_each_entry(cache, &info->block_groups[index], list) {
5461 spin_lock(&cache->lock);
5462 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5463 "%llu pinned %llu reserved\n",
5464 (unsigned long long)cache->key.objectid,
5465 (unsigned long long)cache->key.offset,
5466 (unsigned long long)btrfs_block_group_used(&cache->item),
5467 (unsigned long long)cache->pinned,
5468 (unsigned long long)cache->reserved);
5469 btrfs_dump_free_space(cache, bytes);
5470 spin_unlock(&cache->lock);
5472 if (++index < BTRFS_NR_RAID_TYPES)
5474 up_read(&info->groups_sem);
5477 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5478 struct btrfs_root *root,
5479 u64 num_bytes, u64 min_alloc_size,
5480 u64 empty_size, u64 hint_byte,
5481 u64 search_end, struct btrfs_key *ins,
5485 u64 search_start = 0;
5487 data = btrfs_get_alloc_profile(root, data);
5490 * the only place that sets empty_size is btrfs_realloc_node, which
5491 * is not called recursively on allocations
5493 if (empty_size || root->ref_cows)
5494 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5495 num_bytes + 2 * 1024 * 1024, data,
5496 CHUNK_ALLOC_NO_FORCE);
5498 WARN_ON(num_bytes < root->sectorsize);
5499 ret = find_free_extent(trans, root, num_bytes, empty_size,
5500 search_start, search_end, hint_byte,
5503 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5504 num_bytes = num_bytes >> 1;
5505 num_bytes = num_bytes & ~(root->sectorsize - 1);
5506 num_bytes = max(num_bytes, min_alloc_size);
5507 do_chunk_alloc(trans, root->fs_info->extent_root,
5508 num_bytes, data, CHUNK_ALLOC_FORCE);
5511 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5512 struct btrfs_space_info *sinfo;
5514 sinfo = __find_space_info(root->fs_info, data);
5515 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5516 "wanted %llu\n", (unsigned long long)data,
5517 (unsigned long long)num_bytes);
5518 dump_space_info(sinfo, num_bytes, 1);
5521 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5526 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5528 struct btrfs_block_group_cache *cache;
5531 cache = btrfs_lookup_block_group(root->fs_info, start);
5533 printk(KERN_ERR "Unable to find block group for %llu\n",
5534 (unsigned long long)start);
5538 if (btrfs_test_opt(root, DISCARD))
5539 ret = btrfs_discard_extent(root, start, len, NULL);
5541 btrfs_add_free_space(cache, start, len);
5542 btrfs_update_reserved_bytes(cache, len, 0, 1);
5543 btrfs_put_block_group(cache);
5545 trace_btrfs_reserved_extent_free(root, start, len);
5550 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5551 struct btrfs_root *root,
5552 u64 parent, u64 root_objectid,
5553 u64 flags, u64 owner, u64 offset,
5554 struct btrfs_key *ins, int ref_mod)
5557 struct btrfs_fs_info *fs_info = root->fs_info;
5558 struct btrfs_extent_item *extent_item;
5559 struct btrfs_extent_inline_ref *iref;
5560 struct btrfs_path *path;
5561 struct extent_buffer *leaf;
5566 type = BTRFS_SHARED_DATA_REF_KEY;
5568 type = BTRFS_EXTENT_DATA_REF_KEY;
5570 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5572 path = btrfs_alloc_path();
5576 path->leave_spinning = 1;
5577 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5581 leaf = path->nodes[0];
5582 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5583 struct btrfs_extent_item);
5584 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5585 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5586 btrfs_set_extent_flags(leaf, extent_item,
5587 flags | BTRFS_EXTENT_FLAG_DATA);
5589 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5590 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5592 struct btrfs_shared_data_ref *ref;
5593 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5594 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5595 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5597 struct btrfs_extent_data_ref *ref;
5598 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5599 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5600 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5601 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5602 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5605 btrfs_mark_buffer_dirty(path->nodes[0]);
5606 btrfs_free_path(path);
5608 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5610 printk(KERN_ERR "btrfs update block group failed for %llu "
5611 "%llu\n", (unsigned long long)ins->objectid,
5612 (unsigned long long)ins->offset);
5618 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5619 struct btrfs_root *root,
5620 u64 parent, u64 root_objectid,
5621 u64 flags, struct btrfs_disk_key *key,
5622 int level, struct btrfs_key *ins)
5625 struct btrfs_fs_info *fs_info = root->fs_info;
5626 struct btrfs_extent_item *extent_item;
5627 struct btrfs_tree_block_info *block_info;
5628 struct btrfs_extent_inline_ref *iref;
5629 struct btrfs_path *path;
5630 struct extent_buffer *leaf;
5631 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5633 path = btrfs_alloc_path();
5636 path->leave_spinning = 1;
5637 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5641 leaf = path->nodes[0];
5642 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5643 struct btrfs_extent_item);
5644 btrfs_set_extent_refs(leaf, extent_item, 1);
5645 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5646 btrfs_set_extent_flags(leaf, extent_item,
5647 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5648 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5650 btrfs_set_tree_block_key(leaf, block_info, key);
5651 btrfs_set_tree_block_level(leaf, block_info, level);
5653 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5655 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5656 btrfs_set_extent_inline_ref_type(leaf, iref,
5657 BTRFS_SHARED_BLOCK_REF_KEY);
5658 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5660 btrfs_set_extent_inline_ref_type(leaf, iref,
5661 BTRFS_TREE_BLOCK_REF_KEY);
5662 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5665 btrfs_mark_buffer_dirty(leaf);
5666 btrfs_free_path(path);
5668 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5670 printk(KERN_ERR "btrfs update block group failed for %llu "
5671 "%llu\n", (unsigned long long)ins->objectid,
5672 (unsigned long long)ins->offset);
5678 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5679 struct btrfs_root *root,
5680 u64 root_objectid, u64 owner,
5681 u64 offset, struct btrfs_key *ins)
5685 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5687 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5688 0, root_objectid, owner, offset,
5689 BTRFS_ADD_DELAYED_EXTENT, NULL);
5694 * this is used by the tree logging recovery code. It records that
5695 * an extent has been allocated and makes sure to clear the free
5696 * space cache bits as well
5698 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5699 struct btrfs_root *root,
5700 u64 root_objectid, u64 owner, u64 offset,
5701 struct btrfs_key *ins)
5704 struct btrfs_block_group_cache *block_group;
5705 struct btrfs_caching_control *caching_ctl;
5706 u64 start = ins->objectid;
5707 u64 num_bytes = ins->offset;
5709 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5710 cache_block_group(block_group, trans, NULL, 0);
5711 caching_ctl = get_caching_control(block_group);
5714 BUG_ON(!block_group_cache_done(block_group));
5715 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5718 mutex_lock(&caching_ctl->mutex);
5720 if (start >= caching_ctl->progress) {
5721 ret = add_excluded_extent(root, start, num_bytes);
5723 } else if (start + num_bytes <= caching_ctl->progress) {
5724 ret = btrfs_remove_free_space(block_group,
5728 num_bytes = caching_ctl->progress - start;
5729 ret = btrfs_remove_free_space(block_group,
5733 start = caching_ctl->progress;
5734 num_bytes = ins->objectid + ins->offset -
5735 caching_ctl->progress;
5736 ret = add_excluded_extent(root, start, num_bytes);
5740 mutex_unlock(&caching_ctl->mutex);
5741 put_caching_control(caching_ctl);
5744 ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5746 btrfs_put_block_group(block_group);
5747 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5748 0, owner, offset, ins, 1);
5752 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5753 struct btrfs_root *root,
5754 u64 bytenr, u32 blocksize,
5757 struct extent_buffer *buf;
5759 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5761 return ERR_PTR(-ENOMEM);
5762 btrfs_set_header_generation(buf, trans->transid);
5763 btrfs_set_buffer_lockdep_class(buf, level);
5764 btrfs_tree_lock(buf);
5765 clean_tree_block(trans, root, buf);
5767 btrfs_set_lock_blocking(buf);
5768 btrfs_set_buffer_uptodate(buf);
5770 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5772 * we allow two log transactions at a time, use different
5773 * EXENT bit to differentiate dirty pages.
5775 if (root->log_transid % 2 == 0)
5776 set_extent_dirty(&root->dirty_log_pages, buf->start,
5777 buf->start + buf->len - 1, GFP_NOFS);
5779 set_extent_new(&root->dirty_log_pages, buf->start,
5780 buf->start + buf->len - 1, GFP_NOFS);
5782 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5783 buf->start + buf->len - 1, GFP_NOFS);
5785 trans->blocks_used++;
5786 /* this returns a buffer locked for blocking */
5790 static struct btrfs_block_rsv *
5791 use_block_rsv(struct btrfs_trans_handle *trans,
5792 struct btrfs_root *root, u32 blocksize)
5794 struct btrfs_block_rsv *block_rsv;
5795 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5798 block_rsv = get_block_rsv(trans, root);
5800 if (block_rsv->size == 0) {
5801 ret = reserve_metadata_bytes(trans, root, block_rsv,
5804 * If we couldn't reserve metadata bytes try and use some from
5805 * the global reserve.
5807 if (ret && block_rsv != global_rsv) {
5808 ret = block_rsv_use_bytes(global_rsv, blocksize);
5811 return ERR_PTR(ret);
5813 return ERR_PTR(ret);
5818 ret = block_rsv_use_bytes(block_rsv, blocksize);
5823 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5826 spin_lock(&block_rsv->lock);
5827 block_rsv->size += blocksize;
5828 spin_unlock(&block_rsv->lock);
5830 } else if (ret && block_rsv != global_rsv) {
5831 ret = block_rsv_use_bytes(global_rsv, blocksize);
5837 return ERR_PTR(-ENOSPC);
5840 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5842 block_rsv_add_bytes(block_rsv, blocksize, 0);
5843 block_rsv_release_bytes(block_rsv, NULL, 0);
5847 * finds a free extent and does all the dirty work required for allocation
5848 * returns the key for the extent through ins, and a tree buffer for
5849 * the first block of the extent through buf.
5851 * returns the tree buffer or NULL.
5853 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5854 struct btrfs_root *root, u32 blocksize,
5855 u64 parent, u64 root_objectid,
5856 struct btrfs_disk_key *key, int level,
5857 u64 hint, u64 empty_size)
5859 struct btrfs_key ins;
5860 struct btrfs_block_rsv *block_rsv;
5861 struct extent_buffer *buf;
5866 block_rsv = use_block_rsv(trans, root, blocksize);
5867 if (IS_ERR(block_rsv))
5868 return ERR_CAST(block_rsv);
5870 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5871 empty_size, hint, (u64)-1, &ins, 0);
5873 unuse_block_rsv(block_rsv, blocksize);
5874 return ERR_PTR(ret);
5877 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5879 BUG_ON(IS_ERR(buf));
5881 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5883 parent = ins.objectid;
5884 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5888 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5889 struct btrfs_delayed_extent_op *extent_op;
5890 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5893 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5895 memset(&extent_op->key, 0, sizeof(extent_op->key));
5896 extent_op->flags_to_set = flags;
5897 extent_op->update_key = 1;
5898 extent_op->update_flags = 1;
5899 extent_op->is_data = 0;
5901 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5902 ins.offset, parent, root_objectid,
5903 level, BTRFS_ADD_DELAYED_EXTENT,
5910 struct walk_control {
5911 u64 refs[BTRFS_MAX_LEVEL];
5912 u64 flags[BTRFS_MAX_LEVEL];
5913 struct btrfs_key update_progress;
5923 #define DROP_REFERENCE 1
5924 #define UPDATE_BACKREF 2
5926 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5927 struct btrfs_root *root,
5928 struct walk_control *wc,
5929 struct btrfs_path *path)
5937 struct btrfs_key key;
5938 struct extent_buffer *eb;
5943 if (path->slots[wc->level] < wc->reada_slot) {
5944 wc->reada_count = wc->reada_count * 2 / 3;
5945 wc->reada_count = max(wc->reada_count, 2);
5947 wc->reada_count = wc->reada_count * 3 / 2;
5948 wc->reada_count = min_t(int, wc->reada_count,
5949 BTRFS_NODEPTRS_PER_BLOCK(root));
5952 eb = path->nodes[wc->level];
5953 nritems = btrfs_header_nritems(eb);
5954 blocksize = btrfs_level_size(root, wc->level - 1);
5956 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5957 if (nread >= wc->reada_count)
5961 bytenr = btrfs_node_blockptr(eb, slot);
5962 generation = btrfs_node_ptr_generation(eb, slot);
5964 if (slot == path->slots[wc->level])
5967 if (wc->stage == UPDATE_BACKREF &&
5968 generation <= root->root_key.offset)
5971 /* We don't lock the tree block, it's OK to be racy here */
5972 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5977 if (wc->stage == DROP_REFERENCE) {
5981 if (wc->level == 1 &&
5982 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5984 if (!wc->update_ref ||
5985 generation <= root->root_key.offset)
5987 btrfs_node_key_to_cpu(eb, &key, slot);
5988 ret = btrfs_comp_cpu_keys(&key,
5989 &wc->update_progress);
5993 if (wc->level == 1 &&
5994 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5998 ret = readahead_tree_block(root, bytenr, blocksize,
6004 wc->reada_slot = slot;
6008 * hepler to process tree block while walking down the tree.
6010 * when wc->stage == UPDATE_BACKREF, this function updates
6011 * back refs for pointers in the block.
6013 * NOTE: return value 1 means we should stop walking down.
6015 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6016 struct btrfs_root *root,
6017 struct btrfs_path *path,
6018 struct walk_control *wc, int lookup_info)
6020 int level = wc->level;
6021 struct extent_buffer *eb = path->nodes[level];
6022 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6025 if (wc->stage == UPDATE_BACKREF &&
6026 btrfs_header_owner(eb) != root->root_key.objectid)
6030 * when reference count of tree block is 1, it won't increase
6031 * again. once full backref flag is set, we never clear it.
6034 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6035 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6036 BUG_ON(!path->locks[level]);
6037 ret = btrfs_lookup_extent_info(trans, root,
6042 BUG_ON(wc->refs[level] == 0);
6045 if (wc->stage == DROP_REFERENCE) {
6046 if (wc->refs[level] > 1)
6049 if (path->locks[level] && !wc->keep_locks) {
6050 btrfs_tree_unlock(eb);
6051 path->locks[level] = 0;
6056 /* wc->stage == UPDATE_BACKREF */
6057 if (!(wc->flags[level] & flag)) {
6058 BUG_ON(!path->locks[level]);
6059 ret = btrfs_inc_ref(trans, root, eb, 1);
6061 ret = btrfs_dec_ref(trans, root, eb, 0);
6063 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6066 wc->flags[level] |= flag;
6070 * the block is shared by multiple trees, so it's not good to
6071 * keep the tree lock
6073 if (path->locks[level] && level > 0) {
6074 btrfs_tree_unlock(eb);
6075 path->locks[level] = 0;
6081 * hepler to process tree block pointer.
6083 * when wc->stage == DROP_REFERENCE, this function checks
6084 * reference count of the block pointed to. if the block
6085 * is shared and we need update back refs for the subtree
6086 * rooted at the block, this function changes wc->stage to
6087 * UPDATE_BACKREF. if the block is shared and there is no
6088 * need to update back, this function drops the reference
6091 * NOTE: return value 1 means we should stop walking down.
6093 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6094 struct btrfs_root *root,
6095 struct btrfs_path *path,
6096 struct walk_control *wc, int *lookup_info)
6102 struct btrfs_key key;
6103 struct extent_buffer *next;
6104 int level = wc->level;
6108 generation = btrfs_node_ptr_generation(path->nodes[level],
6109 path->slots[level]);
6111 * if the lower level block was created before the snapshot
6112 * was created, we know there is no need to update back refs
6115 if (wc->stage == UPDATE_BACKREF &&
6116 generation <= root->root_key.offset) {
6121 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6122 blocksize = btrfs_level_size(root, level - 1);
6124 next = btrfs_find_tree_block(root, bytenr, blocksize);
6126 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6131 btrfs_tree_lock(next);
6132 btrfs_set_lock_blocking(next);
6134 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6135 &wc->refs[level - 1],
6136 &wc->flags[level - 1]);
6138 BUG_ON(wc->refs[level - 1] == 0);
6141 if (wc->stage == DROP_REFERENCE) {
6142 if (wc->refs[level - 1] > 1) {
6144 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6147 if (!wc->update_ref ||
6148 generation <= root->root_key.offset)
6151 btrfs_node_key_to_cpu(path->nodes[level], &key,
6152 path->slots[level]);
6153 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6157 wc->stage = UPDATE_BACKREF;
6158 wc->shared_level = level - 1;
6162 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6166 if (!btrfs_buffer_uptodate(next, generation)) {
6167 btrfs_tree_unlock(next);
6168 free_extent_buffer(next);
6174 if (reada && level == 1)
6175 reada_walk_down(trans, root, wc, path);
6176 next = read_tree_block(root, bytenr, blocksize, generation);
6179 btrfs_tree_lock(next);
6180 btrfs_set_lock_blocking(next);
6184 BUG_ON(level != btrfs_header_level(next));
6185 path->nodes[level] = next;
6186 path->slots[level] = 0;
6187 path->locks[level] = 1;
6193 wc->refs[level - 1] = 0;
6194 wc->flags[level - 1] = 0;
6195 if (wc->stage == DROP_REFERENCE) {
6196 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6197 parent = path->nodes[level]->start;
6199 BUG_ON(root->root_key.objectid !=
6200 btrfs_header_owner(path->nodes[level]));
6204 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6205 root->root_key.objectid, level - 1, 0);
6208 btrfs_tree_unlock(next);
6209 free_extent_buffer(next);
6215 * hepler to process tree block while walking up the tree.
6217 * when wc->stage == DROP_REFERENCE, this function drops
6218 * reference count on the block.
6220 * when wc->stage == UPDATE_BACKREF, this function changes
6221 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6222 * to UPDATE_BACKREF previously while processing the block.
6224 * NOTE: return value 1 means we should stop walking up.
6226 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6227 struct btrfs_root *root,
6228 struct btrfs_path *path,
6229 struct walk_control *wc)
6232 int level = wc->level;
6233 struct extent_buffer *eb = path->nodes[level];
6236 if (wc->stage == UPDATE_BACKREF) {
6237 BUG_ON(wc->shared_level < level);
6238 if (level < wc->shared_level)
6241 ret = find_next_key(path, level + 1, &wc->update_progress);
6245 wc->stage = DROP_REFERENCE;
6246 wc->shared_level = -1;
6247 path->slots[level] = 0;
6250 * check reference count again if the block isn't locked.
6251 * we should start walking down the tree again if reference
6254 if (!path->locks[level]) {
6256 btrfs_tree_lock(eb);
6257 btrfs_set_lock_blocking(eb);
6258 path->locks[level] = 1;
6260 ret = btrfs_lookup_extent_info(trans, root,
6265 BUG_ON(wc->refs[level] == 0);
6266 if (wc->refs[level] == 1) {
6267 btrfs_tree_unlock(eb);
6268 path->locks[level] = 0;
6274 /* wc->stage == DROP_REFERENCE */
6275 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6277 if (wc->refs[level] == 1) {
6279 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6280 ret = btrfs_dec_ref(trans, root, eb, 1);
6282 ret = btrfs_dec_ref(trans, root, eb, 0);
6285 /* make block locked assertion in clean_tree_block happy */
6286 if (!path->locks[level] &&
6287 btrfs_header_generation(eb) == trans->transid) {
6288 btrfs_tree_lock(eb);
6289 btrfs_set_lock_blocking(eb);
6290 path->locks[level] = 1;
6292 clean_tree_block(trans, root, eb);
6295 if (eb == root->node) {
6296 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6299 BUG_ON(root->root_key.objectid !=
6300 btrfs_header_owner(eb));
6302 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6303 parent = path->nodes[level + 1]->start;
6305 BUG_ON(root->root_key.objectid !=
6306 btrfs_header_owner(path->nodes[level + 1]));
6309 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6311 wc->refs[level] = 0;
6312 wc->flags[level] = 0;
6316 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6317 struct btrfs_root *root,
6318 struct btrfs_path *path,
6319 struct walk_control *wc)
6321 int level = wc->level;
6322 int lookup_info = 1;
6325 while (level >= 0) {
6326 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6333 if (path->slots[level] >=
6334 btrfs_header_nritems(path->nodes[level]))
6337 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6339 path->slots[level]++;
6348 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6349 struct btrfs_root *root,
6350 struct btrfs_path *path,
6351 struct walk_control *wc, int max_level)
6353 int level = wc->level;
6356 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6357 while (level < max_level && path->nodes[level]) {
6359 if (path->slots[level] + 1 <
6360 btrfs_header_nritems(path->nodes[level])) {
6361 path->slots[level]++;
6364 ret = walk_up_proc(trans, root, path, wc);
6368 if (path->locks[level]) {
6369 btrfs_tree_unlock(path->nodes[level]);
6370 path->locks[level] = 0;
6372 free_extent_buffer(path->nodes[level]);
6373 path->nodes[level] = NULL;
6381 * drop a subvolume tree.
6383 * this function traverses the tree freeing any blocks that only
6384 * referenced by the tree.
6386 * when a shared tree block is found. this function decreases its
6387 * reference count by one. if update_ref is true, this function
6388 * also make sure backrefs for the shared block and all lower level
6389 * blocks are properly updated.
6391 int btrfs_drop_snapshot(struct btrfs_root *root,
6392 struct btrfs_block_rsv *block_rsv, int update_ref)
6394 struct btrfs_path *path;
6395 struct btrfs_trans_handle *trans;
6396 struct btrfs_root *tree_root = root->fs_info->tree_root;
6397 struct btrfs_root_item *root_item = &root->root_item;
6398 struct walk_control *wc;
6399 struct btrfs_key key;
6404 path = btrfs_alloc_path();
6407 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6410 trans = btrfs_start_transaction(tree_root, 0);
6411 BUG_ON(IS_ERR(trans));
6414 trans->block_rsv = block_rsv;
6416 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6417 level = btrfs_header_level(root->node);
6418 path->nodes[level] = btrfs_lock_root_node(root);
6419 btrfs_set_lock_blocking(path->nodes[level]);
6420 path->slots[level] = 0;
6421 path->locks[level] = 1;
6422 memset(&wc->update_progress, 0,
6423 sizeof(wc->update_progress));
6425 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6426 memcpy(&wc->update_progress, &key,
6427 sizeof(wc->update_progress));
6429 level = root_item->drop_level;
6431 path->lowest_level = level;
6432 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6433 path->lowest_level = 0;
6441 * unlock our path, this is safe because only this
6442 * function is allowed to delete this snapshot
6444 btrfs_unlock_up_safe(path, 0);
6446 level = btrfs_header_level(root->node);
6448 btrfs_tree_lock(path->nodes[level]);
6449 btrfs_set_lock_blocking(path->nodes[level]);
6451 ret = btrfs_lookup_extent_info(trans, root,
6452 path->nodes[level]->start,
6453 path->nodes[level]->len,
6457 BUG_ON(wc->refs[level] == 0);
6459 if (level == root_item->drop_level)
6462 btrfs_tree_unlock(path->nodes[level]);
6463 WARN_ON(wc->refs[level] != 1);
6469 wc->shared_level = -1;
6470 wc->stage = DROP_REFERENCE;
6471 wc->update_ref = update_ref;
6473 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6476 ret = walk_down_tree(trans, root, path, wc);
6482 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6489 BUG_ON(wc->stage != DROP_REFERENCE);
6493 if (wc->stage == DROP_REFERENCE) {
6495 btrfs_node_key(path->nodes[level],
6496 &root_item->drop_progress,
6497 path->slots[level]);
6498 root_item->drop_level = level;
6501 BUG_ON(wc->level == 0);
6502 if (btrfs_should_end_transaction(trans, tree_root)) {
6503 ret = btrfs_update_root(trans, tree_root,
6508 btrfs_end_transaction_throttle(trans, tree_root);
6509 trans = btrfs_start_transaction(tree_root, 0);
6510 BUG_ON(IS_ERR(trans));
6512 trans->block_rsv = block_rsv;
6515 btrfs_release_path(root, path);
6518 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6521 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6522 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6526 /* if we fail to delete the orphan item this time
6527 * around, it'll get picked up the next time.
6529 * The most common failure here is just -ENOENT.
6531 btrfs_del_orphan_item(trans, tree_root,
6532 root->root_key.objectid);
6536 if (root->in_radix) {
6537 btrfs_free_fs_root(tree_root->fs_info, root);
6539 free_extent_buffer(root->node);
6540 free_extent_buffer(root->commit_root);
6544 btrfs_end_transaction_throttle(trans, tree_root);
6546 btrfs_free_path(path);
6551 * drop subtree rooted at tree block 'node'.
6553 * NOTE: this function will unlock and release tree block 'node'
6555 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6556 struct btrfs_root *root,
6557 struct extent_buffer *node,
6558 struct extent_buffer *parent)
6560 struct btrfs_path *path;
6561 struct walk_control *wc;
6567 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6569 path = btrfs_alloc_path();
6573 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6575 btrfs_free_path(path);
6579 btrfs_assert_tree_locked(parent);
6580 parent_level = btrfs_header_level(parent);
6581 extent_buffer_get(parent);
6582 path->nodes[parent_level] = parent;
6583 path->slots[parent_level] = btrfs_header_nritems(parent);
6585 btrfs_assert_tree_locked(node);
6586 level = btrfs_header_level(node);
6587 path->nodes[level] = node;
6588 path->slots[level] = 0;
6589 path->locks[level] = 1;
6591 wc->refs[parent_level] = 1;
6592 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6594 wc->shared_level = -1;
6595 wc->stage = DROP_REFERENCE;
6598 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6601 wret = walk_down_tree(trans, root, path, wc);
6607 wret = walk_up_tree(trans, root, path, wc, parent_level);
6615 btrfs_free_path(path);
6620 static unsigned long calc_ra(unsigned long start, unsigned long last,
6623 return min(last, start + nr - 1);
6626 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6631 unsigned long first_index;
6632 unsigned long last_index;
6635 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6636 struct file_ra_state *ra;
6637 struct btrfs_ordered_extent *ordered;
6638 unsigned int total_read = 0;
6639 unsigned int total_dirty = 0;
6642 ra = kzalloc(sizeof(*ra), GFP_NOFS);
6646 mutex_lock(&inode->i_mutex);
6647 first_index = start >> PAGE_CACHE_SHIFT;
6648 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6650 /* make sure the dirty trick played by the caller work */
6651 ret = invalidate_inode_pages2_range(inode->i_mapping,
6652 first_index, last_index);
6656 file_ra_state_init(ra, inode->i_mapping);
6658 for (i = first_index ; i <= last_index; i++) {
6659 if (total_read % ra->ra_pages == 0) {
6660 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6661 calc_ra(i, last_index, ra->ra_pages));
6665 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6667 page = grab_cache_page(inode->i_mapping, i);
6672 if (!PageUptodate(page)) {
6673 btrfs_readpage(NULL, page);
6675 if (!PageUptodate(page)) {
6677 page_cache_release(page);
6682 wait_on_page_writeback(page);
6684 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6685 page_end = page_start + PAGE_CACHE_SIZE - 1;
6686 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6688 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6690 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6692 page_cache_release(page);
6693 btrfs_start_ordered_extent(inode, ordered, 1);
6694 btrfs_put_ordered_extent(ordered);
6697 set_page_extent_mapped(page);
6699 if (i == first_index)
6700 set_extent_bits(io_tree, page_start, page_end,
6701 EXTENT_BOUNDARY, GFP_NOFS);
6702 btrfs_set_extent_delalloc(inode, page_start, page_end);
6704 set_page_dirty(page);
6707 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6709 page_cache_release(page);
6714 mutex_unlock(&inode->i_mutex);
6715 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6719 static noinline int relocate_data_extent(struct inode *reloc_inode,
6720 struct btrfs_key *extent_key,
6723 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6724 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6725 struct extent_map *em;
6726 u64 start = extent_key->objectid - offset;
6727 u64 end = start + extent_key->offset - 1;
6729 em = alloc_extent_map(GFP_NOFS);
6733 em->len = extent_key->offset;
6734 em->block_len = extent_key->offset;
6735 em->block_start = extent_key->objectid;
6736 em->bdev = root->fs_info->fs_devices->latest_bdev;
6737 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6739 /* setup extent map to cheat btrfs_readpage */
6740 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6743 write_lock(&em_tree->lock);
6744 ret = add_extent_mapping(em_tree, em);
6745 write_unlock(&em_tree->lock);
6746 if (ret != -EEXIST) {
6747 free_extent_map(em);
6750 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6752 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6754 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6757 struct btrfs_ref_path {
6759 u64 nodes[BTRFS_MAX_LEVEL];
6761 u64 root_generation;
6768 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6769 u64 new_nodes[BTRFS_MAX_LEVEL];
6772 struct disk_extent {
6783 static int is_cowonly_root(u64 root_objectid)
6785 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6786 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6787 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6788 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6789 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6790 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6795 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6796 struct btrfs_root *extent_root,
6797 struct btrfs_ref_path *ref_path,
6800 struct extent_buffer *leaf;
6801 struct btrfs_path *path;
6802 struct btrfs_extent_ref *ref;
6803 struct btrfs_key key;
6804 struct btrfs_key found_key;
6810 path = btrfs_alloc_path();
6815 ref_path->lowest_level = -1;
6816 ref_path->current_level = -1;
6817 ref_path->shared_level = -1;
6821 level = ref_path->current_level - 1;
6822 while (level >= -1) {
6824 if (level < ref_path->lowest_level)
6828 bytenr = ref_path->nodes[level];
6830 bytenr = ref_path->extent_start;
6831 BUG_ON(bytenr == 0);
6833 parent = ref_path->nodes[level + 1];
6834 ref_path->nodes[level + 1] = 0;
6835 ref_path->current_level = level;
6836 BUG_ON(parent == 0);
6838 key.objectid = bytenr;
6839 key.offset = parent + 1;
6840 key.type = BTRFS_EXTENT_REF_KEY;
6842 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6847 leaf = path->nodes[0];
6848 nritems = btrfs_header_nritems(leaf);
6849 if (path->slots[0] >= nritems) {
6850 ret = btrfs_next_leaf(extent_root, path);
6855 leaf = path->nodes[0];
6858 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6859 if (found_key.objectid == bytenr &&
6860 found_key.type == BTRFS_EXTENT_REF_KEY) {
6861 if (level < ref_path->shared_level)
6862 ref_path->shared_level = level;
6867 btrfs_release_path(extent_root, path);
6870 /* reached lowest level */
6874 level = ref_path->current_level;
6875 while (level < BTRFS_MAX_LEVEL - 1) {
6879 bytenr = ref_path->nodes[level];
6881 bytenr = ref_path->extent_start;
6883 BUG_ON(bytenr == 0);
6885 key.objectid = bytenr;
6887 key.type = BTRFS_EXTENT_REF_KEY;
6889 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6893 leaf = path->nodes[0];
6894 nritems = btrfs_header_nritems(leaf);
6895 if (path->slots[0] >= nritems) {
6896 ret = btrfs_next_leaf(extent_root, path);
6900 /* the extent was freed by someone */
6901 if (ref_path->lowest_level == level)
6903 btrfs_release_path(extent_root, path);
6906 leaf = path->nodes[0];
6909 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6910 if (found_key.objectid != bytenr ||
6911 found_key.type != BTRFS_EXTENT_REF_KEY) {
6912 /* the extent was freed by someone */
6913 if (ref_path->lowest_level == level) {
6917 btrfs_release_path(extent_root, path);
6921 ref = btrfs_item_ptr(leaf, path->slots[0],
6922 struct btrfs_extent_ref);
6923 ref_objectid = btrfs_ref_objectid(leaf, ref);
6924 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6926 level = (int)ref_objectid;
6927 BUG_ON(level >= BTRFS_MAX_LEVEL);
6928 ref_path->lowest_level = level;
6929 ref_path->current_level = level;
6930 ref_path->nodes[level] = bytenr;
6932 WARN_ON(ref_objectid != level);
6935 WARN_ON(level != -1);
6939 if (ref_path->lowest_level == level) {
6940 ref_path->owner_objectid = ref_objectid;
6941 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6945 * the block is tree root or the block isn't in reference
6948 if (found_key.objectid == found_key.offset ||
6949 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6950 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6951 ref_path->root_generation =
6952 btrfs_ref_generation(leaf, ref);
6954 /* special reference from the tree log */
6955 ref_path->nodes[0] = found_key.offset;
6956 ref_path->current_level = 0;
6963 BUG_ON(ref_path->nodes[level] != 0);
6964 ref_path->nodes[level] = found_key.offset;
6965 ref_path->current_level = level;
6968 * the reference was created in the running transaction,
6969 * no need to continue walking up.
6971 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6972 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6973 ref_path->root_generation =
6974 btrfs_ref_generation(leaf, ref);
6979 btrfs_release_path(extent_root, path);
6982 /* reached max tree level, but no tree root found. */
6985 btrfs_free_path(path);
6989 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6990 struct btrfs_root *extent_root,
6991 struct btrfs_ref_path *ref_path,
6994 memset(ref_path, 0, sizeof(*ref_path));
6995 ref_path->extent_start = extent_start;
6997 return __next_ref_path(trans, extent_root, ref_path, 1);
7000 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
7001 struct btrfs_root *extent_root,
7002 struct btrfs_ref_path *ref_path)
7004 return __next_ref_path(trans, extent_root, ref_path, 0);
7007 static noinline int get_new_locations(struct inode *reloc_inode,
7008 struct btrfs_key *extent_key,
7009 u64 offset, int no_fragment,
7010 struct disk_extent **extents,
7013 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
7014 struct btrfs_path *path;
7015 struct btrfs_file_extent_item *fi;
7016 struct extent_buffer *leaf;
7017 struct disk_extent *exts = *extents;
7018 struct btrfs_key found_key;
7023 int max = *nr_extents;
7026 WARN_ON(!no_fragment && *extents);
7029 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
7034 path = btrfs_alloc_path();
7036 if (exts != *extents)
7041 cur_pos = extent_key->objectid - offset;
7042 last_byte = extent_key->objectid + extent_key->offset;
7043 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
7053 leaf = path->nodes[0];
7054 nritems = btrfs_header_nritems(leaf);
7055 if (path->slots[0] >= nritems) {
7056 ret = btrfs_next_leaf(root, path);
7061 leaf = path->nodes[0];
7064 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7065 if (found_key.offset != cur_pos ||
7066 found_key.type != BTRFS_EXTENT_DATA_KEY ||
7067 found_key.objectid != reloc_inode->i_ino)
7070 fi = btrfs_item_ptr(leaf, path->slots[0],
7071 struct btrfs_file_extent_item);
7072 if (btrfs_file_extent_type(leaf, fi) !=
7073 BTRFS_FILE_EXTENT_REG ||
7074 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7078 struct disk_extent *old = exts;
7080 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
7085 memcpy(exts, old, sizeof(*exts) * nr);
7086 if (old != *extents)
7090 exts[nr].disk_bytenr =
7091 btrfs_file_extent_disk_bytenr(leaf, fi);
7092 exts[nr].disk_num_bytes =
7093 btrfs_file_extent_disk_num_bytes(leaf, fi);
7094 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
7095 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7096 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7097 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
7098 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
7099 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
7101 BUG_ON(exts[nr].offset > 0);
7102 BUG_ON(exts[nr].compression || exts[nr].encryption);
7103 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
7105 cur_pos += exts[nr].num_bytes;
7108 if (cur_pos + offset >= last_byte)
7118 BUG_ON(cur_pos + offset > last_byte);
7119 if (cur_pos + offset < last_byte) {
7125 btrfs_free_path(path);
7127 if (exts != *extents)
7136 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
7137 struct btrfs_root *root,
7138 struct btrfs_path *path,
7139 struct btrfs_key *extent_key,
7140 struct btrfs_key *leaf_key,
7141 struct btrfs_ref_path *ref_path,
7142 struct disk_extent *new_extents,
7145 struct extent_buffer *leaf;
7146 struct btrfs_file_extent_item *fi;
7147 struct inode *inode = NULL;
7148 struct btrfs_key key;
7153 u64 search_end = (u64)-1;
7156 int extent_locked = 0;
7160 memcpy(&key, leaf_key, sizeof(key));
7161 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7162 if (key.objectid < ref_path->owner_objectid ||
7163 (key.objectid == ref_path->owner_objectid &&
7164 key.type < BTRFS_EXTENT_DATA_KEY)) {
7165 key.objectid = ref_path->owner_objectid;
7166 key.type = BTRFS_EXTENT_DATA_KEY;
7172 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7176 leaf = path->nodes[0];
7177 nritems = btrfs_header_nritems(leaf);
7179 if (extent_locked && ret > 0) {
7181 * the file extent item was modified by someone
7182 * before the extent got locked.
7184 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7185 lock_end, GFP_NOFS);
7189 if (path->slots[0] >= nritems) {
7190 if (++nr_scaned > 2)
7193 BUG_ON(extent_locked);
7194 ret = btrfs_next_leaf(root, path);
7199 leaf = path->nodes[0];
7200 nritems = btrfs_header_nritems(leaf);
7203 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7205 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7206 if ((key.objectid > ref_path->owner_objectid) ||
7207 (key.objectid == ref_path->owner_objectid &&
7208 key.type > BTRFS_EXTENT_DATA_KEY) ||
7209 key.offset >= search_end)
7213 if (inode && key.objectid != inode->i_ino) {
7214 BUG_ON(extent_locked);
7215 btrfs_release_path(root, path);
7216 mutex_unlock(&inode->i_mutex);
7222 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7227 fi = btrfs_item_ptr(leaf, path->slots[0],
7228 struct btrfs_file_extent_item);
7229 extent_type = btrfs_file_extent_type(leaf, fi);
7230 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7231 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7232 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7233 extent_key->objectid)) {
7239 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7240 ext_offset = btrfs_file_extent_offset(leaf, fi);
7242 if (search_end == (u64)-1) {
7243 search_end = key.offset - ext_offset +
7244 btrfs_file_extent_ram_bytes(leaf, fi);
7247 if (!extent_locked) {
7248 lock_start = key.offset;
7249 lock_end = lock_start + num_bytes - 1;
7251 if (lock_start > key.offset ||
7252 lock_end + 1 < key.offset + num_bytes) {
7253 unlock_extent(&BTRFS_I(inode)->io_tree,
7254 lock_start, lock_end, GFP_NOFS);
7260 btrfs_release_path(root, path);
7262 inode = btrfs_iget_locked(root->fs_info->sb,
7263 key.objectid, root);
7264 if (inode->i_state & I_NEW) {
7265 BTRFS_I(inode)->root = root;
7266 BTRFS_I(inode)->location.objectid =
7268 BTRFS_I(inode)->location.type =
7269 BTRFS_INODE_ITEM_KEY;
7270 BTRFS_I(inode)->location.offset = 0;
7271 btrfs_read_locked_inode(inode);
7272 unlock_new_inode(inode);
7275 * some code call btrfs_commit_transaction while
7276 * holding the i_mutex, so we can't use mutex_lock
7279 if (is_bad_inode(inode) ||
7280 !mutex_trylock(&inode->i_mutex)) {
7283 key.offset = (u64)-1;
7288 if (!extent_locked) {
7289 struct btrfs_ordered_extent *ordered;
7291 btrfs_release_path(root, path);
7293 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7294 lock_end, GFP_NOFS);
7295 ordered = btrfs_lookup_first_ordered_extent(inode,
7298 ordered->file_offset <= lock_end &&
7299 ordered->file_offset + ordered->len > lock_start) {
7300 unlock_extent(&BTRFS_I(inode)->io_tree,
7301 lock_start, lock_end, GFP_NOFS);
7302 btrfs_start_ordered_extent(inode, ordered, 1);
7303 btrfs_put_ordered_extent(ordered);
7304 key.offset += num_bytes;
7308 btrfs_put_ordered_extent(ordered);
7314 if (nr_extents == 1) {
7315 /* update extent pointer in place */
7316 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7317 new_extents[0].disk_bytenr);
7318 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7319 new_extents[0].disk_num_bytes);
7320 btrfs_mark_buffer_dirty(leaf);
7322 btrfs_drop_extent_cache(inode, key.offset,
7323 key.offset + num_bytes - 1, 0);
7325 ret = btrfs_inc_extent_ref(trans, root,
7326 new_extents[0].disk_bytenr,
7327 new_extents[0].disk_num_bytes,
7329 root->root_key.objectid,
7334 ret = btrfs_free_extent(trans, root,
7335 extent_key->objectid,
7338 btrfs_header_owner(leaf),
7339 btrfs_header_generation(leaf),
7343 btrfs_release_path(root, path);
7344 key.offset += num_bytes;
7352 * drop old extent pointer at first, then insert the
7353 * new pointers one bye one
7355 btrfs_release_path(root, path);
7356 ret = btrfs_drop_extents(trans, root, inode, key.offset,
7357 key.offset + num_bytes,
7358 key.offset, &alloc_hint);
7361 for (i = 0; i < nr_extents; i++) {
7362 if (ext_offset >= new_extents[i].num_bytes) {
7363 ext_offset -= new_extents[i].num_bytes;
7366 extent_len = min(new_extents[i].num_bytes -
7367 ext_offset, num_bytes);
7369 ret = btrfs_insert_empty_item(trans, root,
7374 leaf = path->nodes[0];
7375 fi = btrfs_item_ptr(leaf, path->slots[0],
7376 struct btrfs_file_extent_item);
7377 btrfs_set_file_extent_generation(leaf, fi,
7379 btrfs_set_file_extent_type(leaf, fi,
7380 BTRFS_FILE_EXTENT_REG);
7381 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7382 new_extents[i].disk_bytenr);
7383 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7384 new_extents[i].disk_num_bytes);
7385 btrfs_set_file_extent_ram_bytes(leaf, fi,
7386 new_extents[i].ram_bytes);
7388 btrfs_set_file_extent_compression(leaf, fi,
7389 new_extents[i].compression);
7390 btrfs_set_file_extent_encryption(leaf, fi,
7391 new_extents[i].encryption);
7392 btrfs_set_file_extent_other_encoding(leaf, fi,
7393 new_extents[i].other_encoding);
7395 btrfs_set_file_extent_num_bytes(leaf, fi,
7397 ext_offset += new_extents[i].offset;
7398 btrfs_set_file_extent_offset(leaf, fi,
7400 btrfs_mark_buffer_dirty(leaf);
7402 btrfs_drop_extent_cache(inode, key.offset,
7403 key.offset + extent_len - 1, 0);
7405 ret = btrfs_inc_extent_ref(trans, root,
7406 new_extents[i].disk_bytenr,
7407 new_extents[i].disk_num_bytes,
7409 root->root_key.objectid,
7410 trans->transid, key.objectid);
7412 btrfs_release_path(root, path);
7414 inode_add_bytes(inode, extent_len);
7417 num_bytes -= extent_len;
7418 key.offset += extent_len;
7423 BUG_ON(i >= nr_extents);
7427 if (extent_locked) {
7428 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7429 lock_end, GFP_NOFS);
7433 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7434 key.offset >= search_end)
7441 btrfs_release_path(root, path);
7443 mutex_unlock(&inode->i_mutex);
7444 if (extent_locked) {
7445 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7446 lock_end, GFP_NOFS);
7453 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7454 struct btrfs_root *root,
7455 struct extent_buffer *buf, u64 orig_start)
7460 BUG_ON(btrfs_header_generation(buf) != trans->transid);
7461 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7463 level = btrfs_header_level(buf);
7465 struct btrfs_leaf_ref *ref;
7466 struct btrfs_leaf_ref *orig_ref;
7468 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7472 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7474 btrfs_free_leaf_ref(root, orig_ref);
7478 ref->nritems = orig_ref->nritems;
7479 memcpy(ref->extents, orig_ref->extents,
7480 sizeof(ref->extents[0]) * ref->nritems);
7482 btrfs_free_leaf_ref(root, orig_ref);
7484 ref->root_gen = trans->transid;
7485 ref->bytenr = buf->start;
7486 ref->owner = btrfs_header_owner(buf);
7487 ref->generation = btrfs_header_generation(buf);
7489 ret = btrfs_add_leaf_ref(root, ref, 0);
7491 btrfs_free_leaf_ref(root, ref);
7496 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7497 struct extent_buffer *leaf,
7498 struct btrfs_block_group_cache *group,
7499 struct btrfs_root *target_root)
7501 struct btrfs_key key;
7502 struct inode *inode = NULL;
7503 struct btrfs_file_extent_item *fi;
7504 struct extent_state *cached_state = NULL;
7506 u64 skip_objectid = 0;
7510 nritems = btrfs_header_nritems(leaf);
7511 for (i = 0; i < nritems; i++) {
7512 btrfs_item_key_to_cpu(leaf, &key, i);
7513 if (key.objectid == skip_objectid ||
7514 key.type != BTRFS_EXTENT_DATA_KEY)
7516 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7517 if (btrfs_file_extent_type(leaf, fi) ==
7518 BTRFS_FILE_EXTENT_INLINE)
7520 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7522 if (!inode || inode->i_ino != key.objectid) {
7524 inode = btrfs_ilookup(target_root->fs_info->sb,
7525 key.objectid, target_root, 1);
7528 skip_objectid = key.objectid;
7531 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7533 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7534 key.offset + num_bytes - 1, 0, &cached_state,
7536 btrfs_drop_extent_cache(inode, key.offset,
7537 key.offset + num_bytes - 1, 1);
7538 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7539 key.offset + num_bytes - 1, &cached_state,
7547 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7548 struct btrfs_root *root,
7549 struct extent_buffer *leaf,
7550 struct btrfs_block_group_cache *group,
7551 struct inode *reloc_inode)
7553 struct btrfs_key key;
7554 struct btrfs_key extent_key;
7555 struct btrfs_file_extent_item *fi;
7556 struct btrfs_leaf_ref *ref;
7557 struct disk_extent *new_extent;
7566 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7570 ref = btrfs_lookup_leaf_ref(root, leaf->start);
7574 nritems = btrfs_header_nritems(leaf);
7575 for (i = 0; i < nritems; i++) {
7576 btrfs_item_key_to_cpu(leaf, &key, i);
7577 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7579 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7580 if (btrfs_file_extent_type(leaf, fi) ==
7581 BTRFS_FILE_EXTENT_INLINE)
7583 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7584 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7589 if (bytenr >= group->key.objectid + group->key.offset ||
7590 bytenr + num_bytes <= group->key.objectid)
7593 extent_key.objectid = bytenr;
7594 extent_key.offset = num_bytes;
7595 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7597 ret = get_new_locations(reloc_inode, &extent_key,
7598 group->key.objectid, 1,
7599 &new_extent, &nr_extent);
7604 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7605 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7606 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7607 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7609 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7610 new_extent->disk_bytenr);
7611 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7612 new_extent->disk_num_bytes);
7613 btrfs_mark_buffer_dirty(leaf);
7615 ret = btrfs_inc_extent_ref(trans, root,
7616 new_extent->disk_bytenr,
7617 new_extent->disk_num_bytes,
7619 root->root_key.objectid,
7620 trans->transid, key.objectid);
7623 ret = btrfs_free_extent(trans, root,
7624 bytenr, num_bytes, leaf->start,
7625 btrfs_header_owner(leaf),
7626 btrfs_header_generation(leaf),
7632 BUG_ON(ext_index + 1 != ref->nritems);
7633 btrfs_free_leaf_ref(root, ref);
7637 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7638 struct btrfs_root *root)
7640 struct btrfs_root *reloc_root;
7643 if (root->reloc_root) {
7644 reloc_root = root->reloc_root;
7645 root->reloc_root = NULL;
7646 list_add(&reloc_root->dead_list,
7647 &root->fs_info->dead_reloc_roots);
7649 btrfs_set_root_bytenr(&reloc_root->root_item,
7650 reloc_root->node->start);
7651 btrfs_set_root_level(&root->root_item,
7652 btrfs_header_level(reloc_root->node));
7653 memset(&reloc_root->root_item.drop_progress, 0,
7654 sizeof(struct btrfs_disk_key));
7655 reloc_root->root_item.drop_level = 0;
7657 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7658 &reloc_root->root_key,
7659 &reloc_root->root_item);
7665 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7667 struct btrfs_trans_handle *trans;
7668 struct btrfs_root *reloc_root;
7669 struct btrfs_root *prev_root = NULL;
7670 struct list_head dead_roots;
7674 INIT_LIST_HEAD(&dead_roots);
7675 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7677 while (!list_empty(&dead_roots)) {
7678 reloc_root = list_entry(dead_roots.prev,
7679 struct btrfs_root, dead_list);
7680 list_del_init(&reloc_root->dead_list);
7682 BUG_ON(reloc_root->commit_root != NULL);
7684 trans = btrfs_join_transaction(root);
7685 BUG_ON(IS_ERR(trans));
7687 mutex_lock(&root->fs_info->drop_mutex);
7688 ret = btrfs_drop_snapshot(trans, reloc_root);
7691 mutex_unlock(&root->fs_info->drop_mutex);
7693 nr = trans->blocks_used;
7694 ret = btrfs_end_transaction(trans, root);
7696 btrfs_btree_balance_dirty(root, nr);
7699 free_extent_buffer(reloc_root->node);
7701 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7702 &reloc_root->root_key);
7704 mutex_unlock(&root->fs_info->drop_mutex);
7706 nr = trans->blocks_used;
7707 ret = btrfs_end_transaction(trans, root);
7709 btrfs_btree_balance_dirty(root, nr);
7712 prev_root = reloc_root;
7715 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7721 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7723 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7727 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7729 struct btrfs_root *reloc_root;
7730 struct btrfs_trans_handle *trans;
7731 struct btrfs_key location;
7735 mutex_lock(&root->fs_info->tree_reloc_mutex);
7736 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7738 found = !list_empty(&root->fs_info->dead_reloc_roots);
7739 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7742 trans = btrfs_start_transaction(root, 1);
7743 BUG_ON(IS_ERR(trans));
7744 ret = btrfs_commit_transaction(trans, root);
7748 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7749 location.offset = (u64)-1;
7750 location.type = BTRFS_ROOT_ITEM_KEY;
7752 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7753 BUG_ON(!reloc_root);
7754 ret = btrfs_orphan_cleanup(reloc_root);
7759 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7760 struct btrfs_root *root)
7762 struct btrfs_root *reloc_root;
7763 struct extent_buffer *eb;
7764 struct btrfs_root_item *root_item;
7765 struct btrfs_key root_key;
7768 BUG_ON(!root->ref_cows);
7769 if (root->reloc_root)
7772 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7776 ret = btrfs_copy_root(trans, root, root->commit_root,
7777 &eb, BTRFS_TREE_RELOC_OBJECTID);
7780 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7781 root_key.offset = root->root_key.objectid;
7782 root_key.type = BTRFS_ROOT_ITEM_KEY;
7784 memcpy(root_item, &root->root_item, sizeof(root_item));
7785 btrfs_set_root_refs(root_item, 0);
7786 btrfs_set_root_bytenr(root_item, eb->start);
7787 btrfs_set_root_level(root_item, btrfs_header_level(eb));
7788 btrfs_set_root_generation(root_item, trans->transid);
7790 btrfs_tree_unlock(eb);
7791 free_extent_buffer(eb);
7793 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7794 &root_key, root_item);
7798 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7800 BUG_ON(IS_ERR(reloc_root));
7801 reloc_root->last_trans = trans->transid;
7802 reloc_root->commit_root = NULL;
7803 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7805 root->reloc_root = reloc_root;
7810 * Core function of space balance.
7812 * The idea is using reloc trees to relocate tree blocks in reference
7813 * counted roots. There is one reloc tree for each subvol, and all
7814 * reloc trees share same root key objectid. Reloc trees are snapshots
7815 * of the latest committed roots of subvols (root->commit_root).
7817 * To relocate a tree block referenced by a subvol, there are two steps.
7818 * COW the block through subvol's reloc tree, then update block pointer
7819 * in the subvol to point to the new block. Since all reloc trees share
7820 * same root key objectid, doing special handing for tree blocks owned
7821 * by them is easy. Once a tree block has been COWed in one reloc tree,
7822 * we can use the resulting new block directly when the same block is
7823 * required to COW again through other reloc trees. By this way, relocated
7824 * tree blocks are shared between reloc trees, so they are also shared
7827 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7828 struct btrfs_root *root,
7829 struct btrfs_path *path,
7830 struct btrfs_key *first_key,
7831 struct btrfs_ref_path *ref_path,
7832 struct btrfs_block_group_cache *group,
7833 struct inode *reloc_inode)
7835 struct btrfs_root *reloc_root;
7836 struct extent_buffer *eb = NULL;
7837 struct btrfs_key *keys;
7841 int lowest_level = 0;
7844 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7845 lowest_level = ref_path->owner_objectid;
7847 if (!root->ref_cows) {
7848 path->lowest_level = lowest_level;
7849 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7851 path->lowest_level = 0;
7852 btrfs_release_path(root, path);
7856 mutex_lock(&root->fs_info->tree_reloc_mutex);
7857 ret = init_reloc_tree(trans, root);
7859 reloc_root = root->reloc_root;
7861 shared_level = ref_path->shared_level;
7862 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7864 keys = ref_path->node_keys;
7865 nodes = ref_path->new_nodes;
7866 memset(&keys[shared_level + 1], 0,
7867 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7868 memset(&nodes[shared_level + 1], 0,
7869 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7871 if (nodes[lowest_level] == 0) {
7872 path->lowest_level = lowest_level;
7873 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7876 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7877 eb = path->nodes[level];
7878 if (!eb || eb == reloc_root->node)
7880 nodes[level] = eb->start;
7882 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7884 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7887 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7888 eb = path->nodes[0];
7889 ret = replace_extents_in_leaf(trans, reloc_root, eb,
7890 group, reloc_inode);
7893 btrfs_release_path(reloc_root, path);
7895 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7901 * replace tree blocks in the fs tree with tree blocks in
7904 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7907 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7908 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7911 extent_buffer_get(path->nodes[0]);
7912 eb = path->nodes[0];
7913 btrfs_release_path(reloc_root, path);
7914 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7916 free_extent_buffer(eb);
7919 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7920 path->lowest_level = 0;
7924 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7925 struct btrfs_root *root,
7926 struct btrfs_path *path,
7927 struct btrfs_key *first_key,
7928 struct btrfs_ref_path *ref_path)
7932 ret = relocate_one_path(trans, root, path, first_key,
7933 ref_path, NULL, NULL);
7939 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7940 struct btrfs_root *extent_root,
7941 struct btrfs_path *path,
7942 struct btrfs_key *extent_key)
7946 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7949 ret = btrfs_del_item(trans, extent_root, path);
7951 btrfs_release_path(extent_root, path);
7955 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7956 struct btrfs_ref_path *ref_path)
7958 struct btrfs_key root_key;
7960 root_key.objectid = ref_path->root_objectid;
7961 root_key.type = BTRFS_ROOT_ITEM_KEY;
7962 if (is_cowonly_root(ref_path->root_objectid))
7963 root_key.offset = 0;
7965 root_key.offset = (u64)-1;
7967 return btrfs_read_fs_root_no_name(fs_info, &root_key);
7970 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7971 struct btrfs_path *path,
7972 struct btrfs_key *extent_key,
7973 struct btrfs_block_group_cache *group,
7974 struct inode *reloc_inode, int pass)
7976 struct btrfs_trans_handle *trans;
7977 struct btrfs_root *found_root;
7978 struct btrfs_ref_path *ref_path = NULL;
7979 struct disk_extent *new_extents = NULL;
7984 struct btrfs_key first_key;
7988 trans = btrfs_start_transaction(extent_root, 1);
7989 BUG_ON(IS_ERR(trans));
7991 if (extent_key->objectid == 0) {
7992 ret = del_extent_zero(trans, extent_root, path, extent_key);
7996 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
8002 for (loops = 0; ; loops++) {
8004 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
8005 extent_key->objectid);
8007 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
8014 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
8015 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
8018 found_root = read_ref_root(extent_root->fs_info, ref_path);
8019 BUG_ON(!found_root);
8021 * for reference counted tree, only process reference paths
8022 * rooted at the latest committed root.
8024 if (found_root->ref_cows &&
8025 ref_path->root_generation != found_root->root_key.offset)
8028 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8031 * copy data extents to new locations
8033 u64 group_start = group->key.objectid;
8034 ret = relocate_data_extent(reloc_inode,
8043 level = ref_path->owner_objectid;
8046 if (prev_block != ref_path->nodes[level]) {
8047 struct extent_buffer *eb;
8048 u64 block_start = ref_path->nodes[level];
8049 u64 block_size = btrfs_level_size(found_root, level);
8051 eb = read_tree_block(found_root, block_start,
8057 btrfs_tree_lock(eb);
8058 BUG_ON(level != btrfs_header_level(eb));
8061 btrfs_item_key_to_cpu(eb, &first_key, 0);
8063 btrfs_node_key_to_cpu(eb, &first_key, 0);
8065 btrfs_tree_unlock(eb);
8066 free_extent_buffer(eb);
8067 prev_block = block_start;
8070 mutex_lock(&extent_root->fs_info->trans_mutex);
8071 btrfs_record_root_in_trans(found_root);
8072 mutex_unlock(&extent_root->fs_info->trans_mutex);
8073 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8075 * try to update data extent references while
8076 * keeping metadata shared between snapshots.
8079 ret = relocate_one_path(trans, found_root,
8080 path, &first_key, ref_path,
8081 group, reloc_inode);
8087 * use fallback method to process the remaining
8091 u64 group_start = group->key.objectid;
8092 new_extents = kmalloc(sizeof(*new_extents),
8099 ret = get_new_locations(reloc_inode,
8107 ret = replace_one_extent(trans, found_root,
8109 &first_key, ref_path,
8110 new_extents, nr_extents);
8112 ret = relocate_tree_block(trans, found_root, path,
8113 &first_key, ref_path);
8120 btrfs_end_transaction(trans, extent_root);
8127 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8130 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8131 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8134 * we add in the count of missing devices because we want
8135 * to make sure that any RAID levels on a degraded FS
8136 * continue to be honored.
8138 num_devices = root->fs_info->fs_devices->rw_devices +
8139 root->fs_info->fs_devices->missing_devices;
8141 if (num_devices == 1) {
8142 stripped |= BTRFS_BLOCK_GROUP_DUP;
8143 stripped = flags & ~stripped;
8145 /* turn raid0 into single device chunks */
8146 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8149 /* turn mirroring into duplication */
8150 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8151 BTRFS_BLOCK_GROUP_RAID10))
8152 return stripped | BTRFS_BLOCK_GROUP_DUP;
8155 /* they already had raid on here, just return */
8156 if (flags & stripped)
8159 stripped |= BTRFS_BLOCK_GROUP_DUP;
8160 stripped = flags & ~stripped;
8162 /* switch duplicated blocks with raid1 */
8163 if (flags & BTRFS_BLOCK_GROUP_DUP)
8164 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8166 /* turn single device chunks into raid0 */
8167 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8172 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8174 struct btrfs_space_info *sinfo = cache->space_info;
8181 spin_lock(&sinfo->lock);
8182 spin_lock(&cache->lock);
8183 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8184 cache->bytes_super - btrfs_block_group_used(&cache->item);
8186 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8187 sinfo->bytes_may_use + sinfo->bytes_readonly +
8188 cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8189 sinfo->bytes_readonly += num_bytes;
8190 sinfo->bytes_reserved += cache->reserved_pinned;
8191 cache->reserved_pinned = 0;
8196 spin_unlock(&cache->lock);
8197 spin_unlock(&sinfo->lock);
8201 int btrfs_set_block_group_ro(struct btrfs_root *root,
8202 struct btrfs_block_group_cache *cache)
8205 struct btrfs_trans_handle *trans;
8211 trans = btrfs_join_transaction(root);
8212 BUG_ON(IS_ERR(trans));
8214 alloc_flags = update_block_group_flags(root, cache->flags);
8215 if (alloc_flags != cache->flags)
8216 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8219 ret = set_block_group_ro(cache);
8222 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8223 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8227 ret = set_block_group_ro(cache);
8229 btrfs_end_transaction(trans, root);
8233 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8234 struct btrfs_root *root, u64 type)
8236 u64 alloc_flags = get_alloc_profile(root, type);
8237 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8242 * helper to account the unused space of all the readonly block group in the
8243 * list. takes mirrors into account.
8245 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8247 struct btrfs_block_group_cache *block_group;
8251 list_for_each_entry(block_group, groups_list, list) {
8252 spin_lock(&block_group->lock);
8254 if (!block_group->ro) {
8255 spin_unlock(&block_group->lock);
8259 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8260 BTRFS_BLOCK_GROUP_RAID10 |
8261 BTRFS_BLOCK_GROUP_DUP))
8266 free_bytes += (block_group->key.offset -
8267 btrfs_block_group_used(&block_group->item)) *
8270 spin_unlock(&block_group->lock);
8277 * helper to account the unused space of all the readonly block group in the
8278 * space_info. takes mirrors into account.
8280 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8285 spin_lock(&sinfo->lock);
8287 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8288 if (!list_empty(&sinfo->block_groups[i]))
8289 free_bytes += __btrfs_get_ro_block_group_free_space(
8290 &sinfo->block_groups[i]);
8292 spin_unlock(&sinfo->lock);
8297 int btrfs_set_block_group_rw(struct btrfs_root *root,
8298 struct btrfs_block_group_cache *cache)
8300 struct btrfs_space_info *sinfo = cache->space_info;
8305 spin_lock(&sinfo->lock);
8306 spin_lock(&cache->lock);
8307 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8308 cache->bytes_super - btrfs_block_group_used(&cache->item);
8309 sinfo->bytes_readonly -= num_bytes;
8311 spin_unlock(&cache->lock);
8312 spin_unlock(&sinfo->lock);
8317 * checks to see if its even possible to relocate this block group.
8319 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8320 * ok to go ahead and try.
8322 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8324 struct btrfs_block_group_cache *block_group;
8325 struct btrfs_space_info *space_info;
8326 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8327 struct btrfs_device *device;
8331 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8333 /* odd, couldn't find the block group, leave it alone */
8337 /* no bytes used, we're good */
8338 if (!btrfs_block_group_used(&block_group->item))
8341 space_info = block_group->space_info;
8342 spin_lock(&space_info->lock);
8344 full = space_info->full;
8347 * if this is the last block group we have in this space, we can't
8348 * relocate it unless we're able to allocate a new chunk below.
8350 * Otherwise, we need to make sure we have room in the space to handle
8351 * all of the extents from this block group. If we can, we're good
8353 if ((space_info->total_bytes != block_group->key.offset) &&
8354 (space_info->bytes_used + space_info->bytes_reserved +
8355 space_info->bytes_pinned + space_info->bytes_readonly +
8356 btrfs_block_group_used(&block_group->item) <
8357 space_info->total_bytes)) {
8358 spin_unlock(&space_info->lock);
8361 spin_unlock(&space_info->lock);
8364 * ok we don't have enough space, but maybe we have free space on our
8365 * devices to allocate new chunks for relocation, so loop through our
8366 * alloc devices and guess if we have enough space. However, if we
8367 * were marked as full, then we know there aren't enough chunks, and we
8374 mutex_lock(&root->fs_info->chunk_mutex);
8375 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8376 u64 min_free = btrfs_block_group_used(&block_group->item);
8380 * check to make sure we can actually find a chunk with enough
8381 * space to fit our block group in.
8383 if (device->total_bytes > device->bytes_used + min_free) {
8384 ret = find_free_dev_extent(NULL, device, min_free,
8391 mutex_unlock(&root->fs_info->chunk_mutex);
8393 btrfs_put_block_group(block_group);
8397 static int find_first_block_group(struct btrfs_root *root,
8398 struct btrfs_path *path, struct btrfs_key *key)
8401 struct btrfs_key found_key;
8402 struct extent_buffer *leaf;
8405 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8410 slot = path->slots[0];
8411 leaf = path->nodes[0];
8412 if (slot >= btrfs_header_nritems(leaf)) {
8413 ret = btrfs_next_leaf(root, path);
8420 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8422 if (found_key.objectid >= key->objectid &&
8423 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8433 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8435 struct btrfs_block_group_cache *block_group;
8439 struct inode *inode;
8441 block_group = btrfs_lookup_first_block_group(info, last);
8442 while (block_group) {
8443 spin_lock(&block_group->lock);
8444 if (block_group->iref)
8446 spin_unlock(&block_group->lock);
8447 block_group = next_block_group(info->tree_root,
8457 inode = block_group->inode;
8458 block_group->iref = 0;
8459 block_group->inode = NULL;
8460 spin_unlock(&block_group->lock);
8462 last = block_group->key.objectid + block_group->key.offset;
8463 btrfs_put_block_group(block_group);
8467 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8469 struct btrfs_block_group_cache *block_group;
8470 struct btrfs_space_info *space_info;
8471 struct btrfs_caching_control *caching_ctl;
8474 down_write(&info->extent_commit_sem);
8475 while (!list_empty(&info->caching_block_groups)) {
8476 caching_ctl = list_entry(info->caching_block_groups.next,
8477 struct btrfs_caching_control, list);
8478 list_del(&caching_ctl->list);
8479 put_caching_control(caching_ctl);
8481 up_write(&info->extent_commit_sem);
8483 spin_lock(&info->block_group_cache_lock);
8484 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8485 block_group = rb_entry(n, struct btrfs_block_group_cache,
8487 rb_erase(&block_group->cache_node,
8488 &info->block_group_cache_tree);
8489 spin_unlock(&info->block_group_cache_lock);
8491 down_write(&block_group->space_info->groups_sem);
8492 list_del(&block_group->list);
8493 up_write(&block_group->space_info->groups_sem);
8495 if (block_group->cached == BTRFS_CACHE_STARTED)
8496 wait_block_group_cache_done(block_group);
8499 * We haven't cached this block group, which means we could
8500 * possibly have excluded extents on this block group.
8502 if (block_group->cached == BTRFS_CACHE_NO)
8503 free_excluded_extents(info->extent_root, block_group);
8505 btrfs_remove_free_space_cache(block_group);
8506 btrfs_put_block_group(block_group);
8508 spin_lock(&info->block_group_cache_lock);
8510 spin_unlock(&info->block_group_cache_lock);
8512 /* now that all the block groups are freed, go through and
8513 * free all the space_info structs. This is only called during
8514 * the final stages of unmount, and so we know nobody is
8515 * using them. We call synchronize_rcu() once before we start,
8516 * just to be on the safe side.
8520 release_global_block_rsv(info);
8522 while(!list_empty(&info->space_info)) {
8523 space_info = list_entry(info->space_info.next,
8524 struct btrfs_space_info,
8526 if (space_info->bytes_pinned > 0 ||
8527 space_info->bytes_reserved > 0) {
8529 dump_space_info(space_info, 0, 0);
8531 list_del(&space_info->list);
8537 static void __link_block_group(struct btrfs_space_info *space_info,
8538 struct btrfs_block_group_cache *cache)
8540 int index = get_block_group_index(cache);
8542 down_write(&space_info->groups_sem);
8543 list_add_tail(&cache->list, &space_info->block_groups[index]);
8544 up_write(&space_info->groups_sem);
8547 int btrfs_read_block_groups(struct btrfs_root *root)
8549 struct btrfs_path *path;
8551 struct btrfs_block_group_cache *cache;
8552 struct btrfs_fs_info *info = root->fs_info;
8553 struct btrfs_space_info *space_info;
8554 struct btrfs_key key;
8555 struct btrfs_key found_key;
8556 struct extent_buffer *leaf;
8560 root = info->extent_root;
8563 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8564 path = btrfs_alloc_path();
8569 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8570 if (cache_gen != 0 &&
8571 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8573 if (btrfs_test_opt(root, CLEAR_CACHE))
8575 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8576 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8579 ret = find_first_block_group(root, path, &key);
8584 leaf = path->nodes[0];
8585 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8586 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8592 atomic_set(&cache->count, 1);
8593 spin_lock_init(&cache->lock);
8594 spin_lock_init(&cache->tree_lock);
8595 cache->fs_info = info;
8596 INIT_LIST_HEAD(&cache->list);
8597 INIT_LIST_HEAD(&cache->cluster_list);
8600 cache->disk_cache_state = BTRFS_DC_CLEAR;
8603 * we only want to have 32k of ram per block group for keeping
8604 * track of free space, and if we pass 1/2 of that we want to
8605 * start converting things over to using bitmaps
8607 cache->extents_thresh = ((1024 * 32) / 2) /
8608 sizeof(struct btrfs_free_space);
8610 read_extent_buffer(leaf, &cache->item,
8611 btrfs_item_ptr_offset(leaf, path->slots[0]),
8612 sizeof(cache->item));
8613 memcpy(&cache->key, &found_key, sizeof(found_key));
8615 key.objectid = found_key.objectid + found_key.offset;
8616 btrfs_release_path(root, path);
8617 cache->flags = btrfs_block_group_flags(&cache->item);
8618 cache->sectorsize = root->sectorsize;
8621 * We need to exclude the super stripes now so that the space
8622 * info has super bytes accounted for, otherwise we'll think
8623 * we have more space than we actually do.
8625 exclude_super_stripes(root, cache);
8628 * check for two cases, either we are full, and therefore
8629 * don't need to bother with the caching work since we won't
8630 * find any space, or we are empty, and we can just add all
8631 * the space in and be done with it. This saves us _alot_ of
8632 * time, particularly in the full case.
8634 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8635 cache->last_byte_to_unpin = (u64)-1;
8636 cache->cached = BTRFS_CACHE_FINISHED;
8637 free_excluded_extents(root, cache);
8638 } else if (btrfs_block_group_used(&cache->item) == 0) {
8639 cache->last_byte_to_unpin = (u64)-1;
8640 cache->cached = BTRFS_CACHE_FINISHED;
8641 add_new_free_space(cache, root->fs_info,
8643 found_key.objectid +
8645 free_excluded_extents(root, cache);
8648 ret = update_space_info(info, cache->flags, found_key.offset,
8649 btrfs_block_group_used(&cache->item),
8652 cache->space_info = space_info;
8653 spin_lock(&cache->space_info->lock);
8654 cache->space_info->bytes_readonly += cache->bytes_super;
8655 spin_unlock(&cache->space_info->lock);
8657 __link_block_group(space_info, cache);
8659 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8662 set_avail_alloc_bits(root->fs_info, cache->flags);
8663 if (btrfs_chunk_readonly(root, cache->key.objectid))
8664 set_block_group_ro(cache);
8667 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8668 if (!(get_alloc_profile(root, space_info->flags) &
8669 (BTRFS_BLOCK_GROUP_RAID10 |
8670 BTRFS_BLOCK_GROUP_RAID1 |
8671 BTRFS_BLOCK_GROUP_DUP)))
8674 * avoid allocating from un-mirrored block group if there are
8675 * mirrored block groups.
8677 list_for_each_entry(cache, &space_info->block_groups[3], list)
8678 set_block_group_ro(cache);
8679 list_for_each_entry(cache, &space_info->block_groups[4], list)
8680 set_block_group_ro(cache);
8683 init_global_block_rsv(info);
8686 btrfs_free_path(path);
8690 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8691 struct btrfs_root *root, u64 bytes_used,
8692 u64 type, u64 chunk_objectid, u64 chunk_offset,
8696 struct btrfs_root *extent_root;
8697 struct btrfs_block_group_cache *cache;
8699 extent_root = root->fs_info->extent_root;
8701 root->fs_info->last_trans_log_full_commit = trans->transid;
8703 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8707 cache->key.objectid = chunk_offset;
8708 cache->key.offset = size;
8709 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8710 cache->sectorsize = root->sectorsize;
8711 cache->fs_info = root->fs_info;
8714 * we only want to have 32k of ram per block group for keeping track
8715 * of free space, and if we pass 1/2 of that we want to start
8716 * converting things over to using bitmaps
8718 cache->extents_thresh = ((1024 * 32) / 2) /
8719 sizeof(struct btrfs_free_space);
8720 atomic_set(&cache->count, 1);
8721 spin_lock_init(&cache->lock);
8722 spin_lock_init(&cache->tree_lock);
8723 INIT_LIST_HEAD(&cache->list);
8724 INIT_LIST_HEAD(&cache->cluster_list);
8726 btrfs_set_block_group_used(&cache->item, bytes_used);
8727 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8728 cache->flags = type;
8729 btrfs_set_block_group_flags(&cache->item, type);
8731 cache->last_byte_to_unpin = (u64)-1;
8732 cache->cached = BTRFS_CACHE_FINISHED;
8733 exclude_super_stripes(root, cache);
8735 add_new_free_space(cache, root->fs_info, chunk_offset,
8736 chunk_offset + size);
8738 free_excluded_extents(root, cache);
8740 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8741 &cache->space_info);
8744 spin_lock(&cache->space_info->lock);
8745 cache->space_info->bytes_readonly += cache->bytes_super;
8746 spin_unlock(&cache->space_info->lock);
8748 __link_block_group(cache->space_info, cache);
8750 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8753 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8754 sizeof(cache->item));
8757 set_avail_alloc_bits(extent_root->fs_info, type);
8762 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8763 struct btrfs_root *root, u64 group_start)
8765 struct btrfs_path *path;
8766 struct btrfs_block_group_cache *block_group;
8767 struct btrfs_free_cluster *cluster;
8768 struct btrfs_root *tree_root = root->fs_info->tree_root;
8769 struct btrfs_key key;
8770 struct inode *inode;
8774 root = root->fs_info->extent_root;
8776 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8777 BUG_ON(!block_group);
8778 BUG_ON(!block_group->ro);
8781 * Free the reserved super bytes from this block group before
8784 free_excluded_extents(root, block_group);
8786 memcpy(&key, &block_group->key, sizeof(key));
8787 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8788 BTRFS_BLOCK_GROUP_RAID1 |
8789 BTRFS_BLOCK_GROUP_RAID10))
8794 /* make sure this block group isn't part of an allocation cluster */
8795 cluster = &root->fs_info->data_alloc_cluster;
8796 spin_lock(&cluster->refill_lock);
8797 btrfs_return_cluster_to_free_space(block_group, cluster);
8798 spin_unlock(&cluster->refill_lock);
8801 * make sure this block group isn't part of a metadata
8802 * allocation cluster
8804 cluster = &root->fs_info->meta_alloc_cluster;
8805 spin_lock(&cluster->refill_lock);
8806 btrfs_return_cluster_to_free_space(block_group, cluster);
8807 spin_unlock(&cluster->refill_lock);
8809 path = btrfs_alloc_path();
8812 inode = lookup_free_space_inode(root, block_group, path);
8813 if (!IS_ERR(inode)) {
8814 btrfs_orphan_add(trans, inode);
8816 /* One for the block groups ref */
8817 spin_lock(&block_group->lock);
8818 if (block_group->iref) {
8819 block_group->iref = 0;
8820 block_group->inode = NULL;
8821 spin_unlock(&block_group->lock);
8824 spin_unlock(&block_group->lock);
8826 /* One for our lookup ref */
8830 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8831 key.offset = block_group->key.objectid;
8834 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8838 btrfs_release_path(tree_root, path);
8840 ret = btrfs_del_item(trans, tree_root, path);
8843 btrfs_release_path(tree_root, path);
8846 spin_lock(&root->fs_info->block_group_cache_lock);
8847 rb_erase(&block_group->cache_node,
8848 &root->fs_info->block_group_cache_tree);
8849 spin_unlock(&root->fs_info->block_group_cache_lock);
8851 down_write(&block_group->space_info->groups_sem);
8853 * we must use list_del_init so people can check to see if they
8854 * are still on the list after taking the semaphore
8856 list_del_init(&block_group->list);
8857 up_write(&block_group->space_info->groups_sem);
8859 if (block_group->cached == BTRFS_CACHE_STARTED)
8860 wait_block_group_cache_done(block_group);
8862 btrfs_remove_free_space_cache(block_group);
8864 spin_lock(&block_group->space_info->lock);
8865 block_group->space_info->total_bytes -= block_group->key.offset;
8866 block_group->space_info->bytes_readonly -= block_group->key.offset;
8867 block_group->space_info->disk_total -= block_group->key.offset * factor;
8868 spin_unlock(&block_group->space_info->lock);
8870 memcpy(&key, &block_group->key, sizeof(key));
8872 btrfs_clear_space_info_full(root->fs_info);
8874 btrfs_put_block_group(block_group);
8875 btrfs_put_block_group(block_group);
8877 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8883 ret = btrfs_del_item(trans, root, path);
8885 btrfs_free_path(path);
8889 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8891 struct btrfs_space_info *space_info;
8892 struct btrfs_super_block *disk_super;
8898 disk_super = &fs_info->super_copy;
8899 if (!btrfs_super_root(disk_super))
8902 features = btrfs_super_incompat_flags(disk_super);
8903 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8906 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8907 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8912 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8913 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8915 flags = BTRFS_BLOCK_GROUP_METADATA;
8916 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8920 flags = BTRFS_BLOCK_GROUP_DATA;
8921 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8927 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8929 return unpin_extent_range(root, start, end);
8932 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8933 u64 num_bytes, u64 *actual_bytes)
8935 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8938 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8940 struct btrfs_fs_info *fs_info = root->fs_info;
8941 struct btrfs_block_group_cache *cache = NULL;
8948 cache = btrfs_lookup_block_group(fs_info, range->start);
8951 if (cache->key.objectid >= (range->start + range->len)) {
8952 btrfs_put_block_group(cache);
8956 start = max(range->start, cache->key.objectid);
8957 end = min(range->start + range->len,
8958 cache->key.objectid + cache->key.offset);
8960 if (end - start >= range->minlen) {
8961 if (!block_group_cache_done(cache)) {
8962 ret = cache_block_group(cache, NULL, root, 0);
8964 wait_block_group_cache_done(cache);
8966 ret = btrfs_trim_block_group(cache,
8972 trimmed += group_trimmed;
8974 btrfs_put_block_group(cache);
8979 cache = next_block_group(fs_info->tree_root, cache);
8982 range->len = trimmed;